xref: /freebsd/sys/kern/kern_sig.c (revision f4b37ed0f8b307b1f3f0f630ca725d68f1dff30d)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_compat.h"
41 #include "opt_gzio.h"
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/ctype.h>
46 #include <sys/systm.h>
47 #include <sys/signalvar.h>
48 #include <sys/vnode.h>
49 #include <sys/acct.h>
50 #include <sys/bus.h>
51 #include <sys/capsicum.h>
52 #include <sys/condvar.h>
53 #include <sys/event.h>
54 #include <sys/fcntl.h>
55 #include <sys/imgact.h>
56 #include <sys/kernel.h>
57 #include <sys/ktr.h>
58 #include <sys/ktrace.h>
59 #include <sys/lock.h>
60 #include <sys/malloc.h>
61 #include <sys/mutex.h>
62 #include <sys/refcount.h>
63 #include <sys/namei.h>
64 #include <sys/proc.h>
65 #include <sys/procdesc.h>
66 #include <sys/posix4.h>
67 #include <sys/pioctl.h>
68 #include <sys/racct.h>
69 #include <sys/resourcevar.h>
70 #include <sys/sdt.h>
71 #include <sys/sbuf.h>
72 #include <sys/sleepqueue.h>
73 #include <sys/smp.h>
74 #include <sys/stat.h>
75 #include <sys/sx.h>
76 #include <sys/syscallsubr.h>
77 #include <sys/sysctl.h>
78 #include <sys/sysent.h>
79 #include <sys/syslog.h>
80 #include <sys/sysproto.h>
81 #include <sys/timers.h>
82 #include <sys/unistd.h>
83 #include <sys/wait.h>
84 #include <vm/vm.h>
85 #include <vm/vm_extern.h>
86 #include <vm/uma.h>
87 
88 #include <sys/jail.h>
89 
90 #include <machine/cpu.h>
91 
92 #include <security/audit/audit.h>
93 
94 #define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
95 
96 SDT_PROVIDER_DECLARE(proc);
97 SDT_PROBE_DEFINE3(proc, kernel, , signal__send, "struct thread *",
98     "struct proc *", "int");
99 SDT_PROBE_DEFINE2(proc, kernel, , signal__clear, "int",
100     "ksiginfo_t *");
101 SDT_PROBE_DEFINE3(proc, kernel, , signal__discard,
102     "struct thread *", "struct proc *", "int");
103 
104 static int	coredump(struct thread *);
105 static int	killpg1(struct thread *td, int sig, int pgid, int all,
106 		    ksiginfo_t *ksi);
107 static int	issignal(struct thread *td);
108 static int	sigprop(int sig);
109 static void	tdsigwakeup(struct thread *, int, sig_t, int);
110 static void	sig_suspend_threads(struct thread *, struct proc *, int);
111 static int	filt_sigattach(struct knote *kn);
112 static void	filt_sigdetach(struct knote *kn);
113 static int	filt_signal(struct knote *kn, long hint);
114 static struct thread *sigtd(struct proc *p, int sig, int prop);
115 static void	sigqueue_start(void);
116 
117 static uma_zone_t	ksiginfo_zone = NULL;
118 struct filterops sig_filtops = {
119 	.f_isfd = 0,
120 	.f_attach = filt_sigattach,
121 	.f_detach = filt_sigdetach,
122 	.f_event = filt_signal,
123 };
124 
125 static int	kern_logsigexit = 1;
126 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
127     &kern_logsigexit, 0,
128     "Log processes quitting on abnormal signals to syslog(3)");
129 
130 static int	kern_forcesigexit = 1;
131 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
132     &kern_forcesigexit, 0, "Force trap signal to be handled");
133 
134 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW, 0,
135     "POSIX real time signal");
136 
137 static int	max_pending_per_proc = 128;
138 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
139     &max_pending_per_proc, 0, "Max pending signals per proc");
140 
141 static int	preallocate_siginfo = 1024;
142 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
143     &preallocate_siginfo, 0, "Preallocated signal memory size");
144 
145 static int	signal_overflow = 0;
146 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
147     &signal_overflow, 0, "Number of signals overflew");
148 
149 static int	signal_alloc_fail = 0;
150 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
151     &signal_alloc_fail, 0, "signals failed to be allocated");
152 
153 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
154 
155 /*
156  * Policy -- Can ucred cr1 send SIGIO to process cr2?
157  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
158  * in the right situations.
159  */
160 #define CANSIGIO(cr1, cr2) \
161 	((cr1)->cr_uid == 0 || \
162 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
163 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
164 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
165 	    (cr1)->cr_uid == (cr2)->cr_uid)
166 
167 static int	sugid_coredump;
168 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
169     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
170 
171 static int	capmode_coredump;
172 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
173     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
174 
175 static int	do_coredump = 1;
176 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
177 	&do_coredump, 0, "Enable/Disable coredumps");
178 
179 static int	set_core_nodump_flag = 0;
180 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
181 	0, "Enable setting the NODUMP flag on coredump files");
182 
183 static int	coredump_devctl = 0;
184 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
185 	0, "Generate a devctl notification when processes coredump");
186 
187 /*
188  * Signal properties and actions.
189  * The array below categorizes the signals and their default actions
190  * according to the following properties:
191  */
192 #define	SA_KILL		0x01		/* terminates process by default */
193 #define	SA_CORE		0x02		/* ditto and coredumps */
194 #define	SA_STOP		0x04		/* suspend process */
195 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
196 #define	SA_IGNORE	0x10		/* ignore by default */
197 #define	SA_CONT		0x20		/* continue if suspended */
198 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
199 
200 static int sigproptbl[NSIG] = {
201 	SA_KILL,			/* SIGHUP */
202 	SA_KILL,			/* SIGINT */
203 	SA_KILL|SA_CORE,		/* SIGQUIT */
204 	SA_KILL|SA_CORE,		/* SIGILL */
205 	SA_KILL|SA_CORE,		/* SIGTRAP */
206 	SA_KILL|SA_CORE,		/* SIGABRT */
207 	SA_KILL|SA_CORE,		/* SIGEMT */
208 	SA_KILL|SA_CORE,		/* SIGFPE */
209 	SA_KILL,			/* SIGKILL */
210 	SA_KILL|SA_CORE,		/* SIGBUS */
211 	SA_KILL|SA_CORE,		/* SIGSEGV */
212 	SA_KILL|SA_CORE,		/* SIGSYS */
213 	SA_KILL,			/* SIGPIPE */
214 	SA_KILL,			/* SIGALRM */
215 	SA_KILL,			/* SIGTERM */
216 	SA_IGNORE,			/* SIGURG */
217 	SA_STOP,			/* SIGSTOP */
218 	SA_STOP|SA_TTYSTOP,		/* SIGTSTP */
219 	SA_IGNORE|SA_CONT,		/* SIGCONT */
220 	SA_IGNORE,			/* SIGCHLD */
221 	SA_STOP|SA_TTYSTOP,		/* SIGTTIN */
222 	SA_STOP|SA_TTYSTOP,		/* SIGTTOU */
223 	SA_IGNORE,			/* SIGIO */
224 	SA_KILL,			/* SIGXCPU */
225 	SA_KILL,			/* SIGXFSZ */
226 	SA_KILL,			/* SIGVTALRM */
227 	SA_KILL,			/* SIGPROF */
228 	SA_IGNORE,			/* SIGWINCH  */
229 	SA_IGNORE,			/* SIGINFO */
230 	SA_KILL,			/* SIGUSR1 */
231 	SA_KILL,			/* SIGUSR2 */
232 };
233 
234 static void reschedule_signals(struct proc *p, sigset_t block, int flags);
235 
236 static void
237 sigqueue_start(void)
238 {
239 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
240 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
241 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
242 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
243 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
244 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
245 }
246 
247 ksiginfo_t *
248 ksiginfo_alloc(int wait)
249 {
250 	int flags;
251 
252 	flags = M_ZERO;
253 	if (! wait)
254 		flags |= M_NOWAIT;
255 	if (ksiginfo_zone != NULL)
256 		return ((ksiginfo_t *)uma_zalloc(ksiginfo_zone, flags));
257 	return (NULL);
258 }
259 
260 void
261 ksiginfo_free(ksiginfo_t *ksi)
262 {
263 	uma_zfree(ksiginfo_zone, ksi);
264 }
265 
266 static __inline int
267 ksiginfo_tryfree(ksiginfo_t *ksi)
268 {
269 	if (!(ksi->ksi_flags & KSI_EXT)) {
270 		uma_zfree(ksiginfo_zone, ksi);
271 		return (1);
272 	}
273 	return (0);
274 }
275 
276 void
277 sigqueue_init(sigqueue_t *list, struct proc *p)
278 {
279 	SIGEMPTYSET(list->sq_signals);
280 	SIGEMPTYSET(list->sq_kill);
281 	TAILQ_INIT(&list->sq_list);
282 	list->sq_proc = p;
283 	list->sq_flags = SQ_INIT;
284 }
285 
286 /*
287  * Get a signal's ksiginfo.
288  * Return:
289  *	0	-	signal not found
290  *	others	-	signal number
291  */
292 static int
293 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
294 {
295 	struct proc *p = sq->sq_proc;
296 	struct ksiginfo *ksi, *next;
297 	int count = 0;
298 
299 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
300 
301 	if (!SIGISMEMBER(sq->sq_signals, signo))
302 		return (0);
303 
304 	if (SIGISMEMBER(sq->sq_kill, signo)) {
305 		count++;
306 		SIGDELSET(sq->sq_kill, signo);
307 	}
308 
309 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
310 		if (ksi->ksi_signo == signo) {
311 			if (count == 0) {
312 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
313 				ksi->ksi_sigq = NULL;
314 				ksiginfo_copy(ksi, si);
315 				if (ksiginfo_tryfree(ksi) && p != NULL)
316 					p->p_pendingcnt--;
317 			}
318 			if (++count > 1)
319 				break;
320 		}
321 	}
322 
323 	if (count <= 1)
324 		SIGDELSET(sq->sq_signals, signo);
325 	si->ksi_signo = signo;
326 	return (signo);
327 }
328 
329 void
330 sigqueue_take(ksiginfo_t *ksi)
331 {
332 	struct ksiginfo *kp;
333 	struct proc	*p;
334 	sigqueue_t	*sq;
335 
336 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
337 		return;
338 
339 	p = sq->sq_proc;
340 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
341 	ksi->ksi_sigq = NULL;
342 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
343 		p->p_pendingcnt--;
344 
345 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
346 	     kp = TAILQ_NEXT(kp, ksi_link)) {
347 		if (kp->ksi_signo == ksi->ksi_signo)
348 			break;
349 	}
350 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo))
351 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
352 }
353 
354 static int
355 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
356 {
357 	struct proc *p = sq->sq_proc;
358 	struct ksiginfo *ksi;
359 	int ret = 0;
360 
361 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
362 
363 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
364 		SIGADDSET(sq->sq_kill, signo);
365 		goto out_set_bit;
366 	}
367 
368 	/* directly insert the ksi, don't copy it */
369 	if (si->ksi_flags & KSI_INS) {
370 		if (si->ksi_flags & KSI_HEAD)
371 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
372 		else
373 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
374 		si->ksi_sigq = sq;
375 		goto out_set_bit;
376 	}
377 
378 	if (__predict_false(ksiginfo_zone == NULL)) {
379 		SIGADDSET(sq->sq_kill, signo);
380 		goto out_set_bit;
381 	}
382 
383 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
384 		signal_overflow++;
385 		ret = EAGAIN;
386 	} else if ((ksi = ksiginfo_alloc(0)) == NULL) {
387 		signal_alloc_fail++;
388 		ret = EAGAIN;
389 	} else {
390 		if (p != NULL)
391 			p->p_pendingcnt++;
392 		ksiginfo_copy(si, ksi);
393 		ksi->ksi_signo = signo;
394 		if (si->ksi_flags & KSI_HEAD)
395 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
396 		else
397 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
398 		ksi->ksi_sigq = sq;
399 	}
400 
401 	if ((si->ksi_flags & KSI_TRAP) != 0 ||
402 	    (si->ksi_flags & KSI_SIGQ) == 0) {
403 		if (ret != 0)
404 			SIGADDSET(sq->sq_kill, signo);
405 		ret = 0;
406 		goto out_set_bit;
407 	}
408 
409 	if (ret != 0)
410 		return (ret);
411 
412 out_set_bit:
413 	SIGADDSET(sq->sq_signals, signo);
414 	return (ret);
415 }
416 
417 void
418 sigqueue_flush(sigqueue_t *sq)
419 {
420 	struct proc *p = sq->sq_proc;
421 	ksiginfo_t *ksi;
422 
423 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
424 
425 	if (p != NULL)
426 		PROC_LOCK_ASSERT(p, MA_OWNED);
427 
428 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
429 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
430 		ksi->ksi_sigq = NULL;
431 		if (ksiginfo_tryfree(ksi) && p != NULL)
432 			p->p_pendingcnt--;
433 	}
434 
435 	SIGEMPTYSET(sq->sq_signals);
436 	SIGEMPTYSET(sq->sq_kill);
437 }
438 
439 static void
440 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
441 {
442 	sigset_t tmp;
443 	struct proc *p1, *p2;
444 	ksiginfo_t *ksi, *next;
445 
446 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
447 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
448 	p1 = src->sq_proc;
449 	p2 = dst->sq_proc;
450 	/* Move siginfo to target list */
451 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
452 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
453 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
454 			if (p1 != NULL)
455 				p1->p_pendingcnt--;
456 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
457 			ksi->ksi_sigq = dst;
458 			if (p2 != NULL)
459 				p2->p_pendingcnt++;
460 		}
461 	}
462 
463 	/* Move pending bits to target list */
464 	tmp = src->sq_kill;
465 	SIGSETAND(tmp, *set);
466 	SIGSETOR(dst->sq_kill, tmp);
467 	SIGSETNAND(src->sq_kill, tmp);
468 
469 	tmp = src->sq_signals;
470 	SIGSETAND(tmp, *set);
471 	SIGSETOR(dst->sq_signals, tmp);
472 	SIGSETNAND(src->sq_signals, tmp);
473 }
474 
475 #if 0
476 static void
477 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
478 {
479 	sigset_t set;
480 
481 	SIGEMPTYSET(set);
482 	SIGADDSET(set, signo);
483 	sigqueue_move_set(src, dst, &set);
484 }
485 #endif
486 
487 static void
488 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
489 {
490 	struct proc *p = sq->sq_proc;
491 	ksiginfo_t *ksi, *next;
492 
493 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
494 
495 	/* Remove siginfo queue */
496 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
497 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
498 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
499 			ksi->ksi_sigq = NULL;
500 			if (ksiginfo_tryfree(ksi) && p != NULL)
501 				p->p_pendingcnt--;
502 		}
503 	}
504 	SIGSETNAND(sq->sq_kill, *set);
505 	SIGSETNAND(sq->sq_signals, *set);
506 }
507 
508 void
509 sigqueue_delete(sigqueue_t *sq, int signo)
510 {
511 	sigset_t set;
512 
513 	SIGEMPTYSET(set);
514 	SIGADDSET(set, signo);
515 	sigqueue_delete_set(sq, &set);
516 }
517 
518 /* Remove a set of signals for a process */
519 static void
520 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
521 {
522 	sigqueue_t worklist;
523 	struct thread *td0;
524 
525 	PROC_LOCK_ASSERT(p, MA_OWNED);
526 
527 	sigqueue_init(&worklist, NULL);
528 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
529 
530 	FOREACH_THREAD_IN_PROC(p, td0)
531 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
532 
533 	sigqueue_flush(&worklist);
534 }
535 
536 void
537 sigqueue_delete_proc(struct proc *p, int signo)
538 {
539 	sigset_t set;
540 
541 	SIGEMPTYSET(set);
542 	SIGADDSET(set, signo);
543 	sigqueue_delete_set_proc(p, &set);
544 }
545 
546 static void
547 sigqueue_delete_stopmask_proc(struct proc *p)
548 {
549 	sigset_t set;
550 
551 	SIGEMPTYSET(set);
552 	SIGADDSET(set, SIGSTOP);
553 	SIGADDSET(set, SIGTSTP);
554 	SIGADDSET(set, SIGTTIN);
555 	SIGADDSET(set, SIGTTOU);
556 	sigqueue_delete_set_proc(p, &set);
557 }
558 
559 /*
560  * Determine signal that should be delivered to thread td, the current
561  * thread, 0 if none.  If there is a pending stop signal with default
562  * action, the process stops in issignal().
563  */
564 int
565 cursig(struct thread *td)
566 {
567 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
568 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
569 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
570 	return (SIGPENDING(td) ? issignal(td) : 0);
571 }
572 
573 /*
574  * Arrange for ast() to handle unmasked pending signals on return to user
575  * mode.  This must be called whenever a signal is added to td_sigqueue or
576  * unmasked in td_sigmask.
577  */
578 void
579 signotify(struct thread *td)
580 {
581 	struct proc *p;
582 
583 	p = td->td_proc;
584 
585 	PROC_LOCK_ASSERT(p, MA_OWNED);
586 
587 	if (SIGPENDING(td)) {
588 		thread_lock(td);
589 		td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING;
590 		thread_unlock(td);
591 	}
592 }
593 
594 int
595 sigonstack(size_t sp)
596 {
597 	struct thread *td = curthread;
598 
599 	return ((td->td_pflags & TDP_ALTSTACK) ?
600 #if defined(COMPAT_43)
601 	    ((td->td_sigstk.ss_size == 0) ?
602 		(td->td_sigstk.ss_flags & SS_ONSTACK) :
603 		((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size))
604 #else
605 	    ((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size)
606 #endif
607 	    : 0);
608 }
609 
610 static __inline int
611 sigprop(int sig)
612 {
613 
614 	if (sig > 0 && sig < NSIG)
615 		return (sigproptbl[_SIG_IDX(sig)]);
616 	return (0);
617 }
618 
619 int
620 sig_ffs(sigset_t *set)
621 {
622 	int i;
623 
624 	for (i = 0; i < _SIG_WORDS; i++)
625 		if (set->__bits[i])
626 			return (ffs(set->__bits[i]) + (i * 32));
627 	return (0);
628 }
629 
630 static bool
631 sigact_flag_test(const struct sigaction *act, int flag)
632 {
633 
634 	/*
635 	 * SA_SIGINFO is reset when signal disposition is set to
636 	 * ignore or default.  Other flags are kept according to user
637 	 * settings.
638 	 */
639 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
640 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
641 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
642 }
643 
644 /*
645  * kern_sigaction
646  * sigaction
647  * freebsd4_sigaction
648  * osigaction
649  */
650 int
651 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
652     struct sigaction *oact, int flags)
653 {
654 	struct sigacts *ps;
655 	struct proc *p = td->td_proc;
656 
657 	if (!_SIG_VALID(sig))
658 		return (EINVAL);
659 	if (act != NULL && act->sa_handler != SIG_DFL &&
660 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
661 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
662 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
663 		return (EINVAL);
664 
665 	PROC_LOCK(p);
666 	ps = p->p_sigacts;
667 	mtx_lock(&ps->ps_mtx);
668 	if (oact) {
669 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
670 		oact->sa_flags = 0;
671 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
672 			oact->sa_flags |= SA_ONSTACK;
673 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
674 			oact->sa_flags |= SA_RESTART;
675 		if (SIGISMEMBER(ps->ps_sigreset, sig))
676 			oact->sa_flags |= SA_RESETHAND;
677 		if (SIGISMEMBER(ps->ps_signodefer, sig))
678 			oact->sa_flags |= SA_NODEFER;
679 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
680 			oact->sa_flags |= SA_SIGINFO;
681 			oact->sa_sigaction =
682 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
683 		} else
684 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
685 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
686 			oact->sa_flags |= SA_NOCLDSTOP;
687 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
688 			oact->sa_flags |= SA_NOCLDWAIT;
689 	}
690 	if (act) {
691 		if ((sig == SIGKILL || sig == SIGSTOP) &&
692 		    act->sa_handler != SIG_DFL) {
693 			mtx_unlock(&ps->ps_mtx);
694 			PROC_UNLOCK(p);
695 			return (EINVAL);
696 		}
697 
698 		/*
699 		 * Change setting atomically.
700 		 */
701 
702 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
703 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
704 		if (sigact_flag_test(act, SA_SIGINFO)) {
705 			ps->ps_sigact[_SIG_IDX(sig)] =
706 			    (__sighandler_t *)act->sa_sigaction;
707 			SIGADDSET(ps->ps_siginfo, sig);
708 		} else {
709 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
710 			SIGDELSET(ps->ps_siginfo, sig);
711 		}
712 		if (!sigact_flag_test(act, SA_RESTART))
713 			SIGADDSET(ps->ps_sigintr, sig);
714 		else
715 			SIGDELSET(ps->ps_sigintr, sig);
716 		if (sigact_flag_test(act, SA_ONSTACK))
717 			SIGADDSET(ps->ps_sigonstack, sig);
718 		else
719 			SIGDELSET(ps->ps_sigonstack, sig);
720 		if (sigact_flag_test(act, SA_RESETHAND))
721 			SIGADDSET(ps->ps_sigreset, sig);
722 		else
723 			SIGDELSET(ps->ps_sigreset, sig);
724 		if (sigact_flag_test(act, SA_NODEFER))
725 			SIGADDSET(ps->ps_signodefer, sig);
726 		else
727 			SIGDELSET(ps->ps_signodefer, sig);
728 		if (sig == SIGCHLD) {
729 			if (act->sa_flags & SA_NOCLDSTOP)
730 				ps->ps_flag |= PS_NOCLDSTOP;
731 			else
732 				ps->ps_flag &= ~PS_NOCLDSTOP;
733 			if (act->sa_flags & SA_NOCLDWAIT) {
734 				/*
735 				 * Paranoia: since SA_NOCLDWAIT is implemented
736 				 * by reparenting the dying child to PID 1 (and
737 				 * trust it to reap the zombie), PID 1 itself
738 				 * is forbidden to set SA_NOCLDWAIT.
739 				 */
740 				if (p->p_pid == 1)
741 					ps->ps_flag &= ~PS_NOCLDWAIT;
742 				else
743 					ps->ps_flag |= PS_NOCLDWAIT;
744 			} else
745 				ps->ps_flag &= ~PS_NOCLDWAIT;
746 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
747 				ps->ps_flag |= PS_CLDSIGIGN;
748 			else
749 				ps->ps_flag &= ~PS_CLDSIGIGN;
750 		}
751 		/*
752 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
753 		 * and for signals set to SIG_DFL where the default is to
754 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
755 		 * have to restart the process.
756 		 */
757 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
758 		    (sigprop(sig) & SA_IGNORE &&
759 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
760 			/* never to be seen again */
761 			sigqueue_delete_proc(p, sig);
762 			if (sig != SIGCONT)
763 				/* easier in psignal */
764 				SIGADDSET(ps->ps_sigignore, sig);
765 			SIGDELSET(ps->ps_sigcatch, sig);
766 		} else {
767 			SIGDELSET(ps->ps_sigignore, sig);
768 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
769 				SIGDELSET(ps->ps_sigcatch, sig);
770 			else
771 				SIGADDSET(ps->ps_sigcatch, sig);
772 		}
773 #ifdef COMPAT_FREEBSD4
774 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
775 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
776 		    (flags & KSA_FREEBSD4) == 0)
777 			SIGDELSET(ps->ps_freebsd4, sig);
778 		else
779 			SIGADDSET(ps->ps_freebsd4, sig);
780 #endif
781 #ifdef COMPAT_43
782 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
783 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
784 		    (flags & KSA_OSIGSET) == 0)
785 			SIGDELSET(ps->ps_osigset, sig);
786 		else
787 			SIGADDSET(ps->ps_osigset, sig);
788 #endif
789 	}
790 	mtx_unlock(&ps->ps_mtx);
791 	PROC_UNLOCK(p);
792 	return (0);
793 }
794 
795 #ifndef _SYS_SYSPROTO_H_
796 struct sigaction_args {
797 	int	sig;
798 	struct	sigaction *act;
799 	struct	sigaction *oact;
800 };
801 #endif
802 int
803 sys_sigaction(td, uap)
804 	struct thread *td;
805 	register struct sigaction_args *uap;
806 {
807 	struct sigaction act, oact;
808 	register struct sigaction *actp, *oactp;
809 	int error;
810 
811 	actp = (uap->act != NULL) ? &act : NULL;
812 	oactp = (uap->oact != NULL) ? &oact : NULL;
813 	if (actp) {
814 		error = copyin(uap->act, actp, sizeof(act));
815 		if (error)
816 			return (error);
817 	}
818 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
819 	if (oactp && !error)
820 		error = copyout(oactp, uap->oact, sizeof(oact));
821 	return (error);
822 }
823 
824 #ifdef COMPAT_FREEBSD4
825 #ifndef _SYS_SYSPROTO_H_
826 struct freebsd4_sigaction_args {
827 	int	sig;
828 	struct	sigaction *act;
829 	struct	sigaction *oact;
830 };
831 #endif
832 int
833 freebsd4_sigaction(td, uap)
834 	struct thread *td;
835 	register struct freebsd4_sigaction_args *uap;
836 {
837 	struct sigaction act, oact;
838 	register struct sigaction *actp, *oactp;
839 	int error;
840 
841 
842 	actp = (uap->act != NULL) ? &act : NULL;
843 	oactp = (uap->oact != NULL) ? &oact : NULL;
844 	if (actp) {
845 		error = copyin(uap->act, actp, sizeof(act));
846 		if (error)
847 			return (error);
848 	}
849 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
850 	if (oactp && !error)
851 		error = copyout(oactp, uap->oact, sizeof(oact));
852 	return (error);
853 }
854 #endif	/* COMAPT_FREEBSD4 */
855 
856 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
857 #ifndef _SYS_SYSPROTO_H_
858 struct osigaction_args {
859 	int	signum;
860 	struct	osigaction *nsa;
861 	struct	osigaction *osa;
862 };
863 #endif
864 int
865 osigaction(td, uap)
866 	struct thread *td;
867 	register struct osigaction_args *uap;
868 {
869 	struct osigaction sa;
870 	struct sigaction nsa, osa;
871 	register struct sigaction *nsap, *osap;
872 	int error;
873 
874 	if (uap->signum <= 0 || uap->signum >= ONSIG)
875 		return (EINVAL);
876 
877 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
878 	osap = (uap->osa != NULL) ? &osa : NULL;
879 
880 	if (nsap) {
881 		error = copyin(uap->nsa, &sa, sizeof(sa));
882 		if (error)
883 			return (error);
884 		nsap->sa_handler = sa.sa_handler;
885 		nsap->sa_flags = sa.sa_flags;
886 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
887 	}
888 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
889 	if (osap && !error) {
890 		sa.sa_handler = osap->sa_handler;
891 		sa.sa_flags = osap->sa_flags;
892 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
893 		error = copyout(&sa, uap->osa, sizeof(sa));
894 	}
895 	return (error);
896 }
897 
898 #if !defined(__i386__)
899 /* Avoid replicating the same stub everywhere */
900 int
901 osigreturn(td, uap)
902 	struct thread *td;
903 	struct osigreturn_args *uap;
904 {
905 
906 	return (nosys(td, (struct nosys_args *)uap));
907 }
908 #endif
909 #endif /* COMPAT_43 */
910 
911 /*
912  * Initialize signal state for process 0;
913  * set to ignore signals that are ignored by default.
914  */
915 void
916 siginit(p)
917 	struct proc *p;
918 {
919 	register int i;
920 	struct sigacts *ps;
921 
922 	PROC_LOCK(p);
923 	ps = p->p_sigacts;
924 	mtx_lock(&ps->ps_mtx);
925 	for (i = 1; i <= NSIG; i++) {
926 		if (sigprop(i) & SA_IGNORE && i != SIGCONT) {
927 			SIGADDSET(ps->ps_sigignore, i);
928 		}
929 	}
930 	mtx_unlock(&ps->ps_mtx);
931 	PROC_UNLOCK(p);
932 }
933 
934 /*
935  * Reset specified signal to the default disposition.
936  */
937 static void
938 sigdflt(struct sigacts *ps, int sig)
939 {
940 
941 	mtx_assert(&ps->ps_mtx, MA_OWNED);
942 	SIGDELSET(ps->ps_sigcatch, sig);
943 	if ((sigprop(sig) & SA_IGNORE) != 0 && sig != SIGCONT)
944 		SIGADDSET(ps->ps_sigignore, sig);
945 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
946 	SIGDELSET(ps->ps_siginfo, sig);
947 }
948 
949 /*
950  * Reset signals for an exec of the specified process.
951  */
952 void
953 execsigs(struct proc *p)
954 {
955 	struct sigacts *ps;
956 	int sig;
957 	struct thread *td;
958 
959 	/*
960 	 * Reset caught signals.  Held signals remain held
961 	 * through td_sigmask (unless they were caught,
962 	 * and are now ignored by default).
963 	 */
964 	PROC_LOCK_ASSERT(p, MA_OWNED);
965 	td = FIRST_THREAD_IN_PROC(p);
966 	ps = p->p_sigacts;
967 	mtx_lock(&ps->ps_mtx);
968 	while (SIGNOTEMPTY(ps->ps_sigcatch)) {
969 		sig = sig_ffs(&ps->ps_sigcatch);
970 		sigdflt(ps, sig);
971 		if ((sigprop(sig) & SA_IGNORE) != 0)
972 			sigqueue_delete_proc(p, sig);
973 	}
974 	/*
975 	 * Reset stack state to the user stack.
976 	 * Clear set of signals caught on the signal stack.
977 	 */
978 	td->td_sigstk.ss_flags = SS_DISABLE;
979 	td->td_sigstk.ss_size = 0;
980 	td->td_sigstk.ss_sp = 0;
981 	td->td_pflags &= ~TDP_ALTSTACK;
982 	/*
983 	 * Reset no zombies if child dies flag as Solaris does.
984 	 */
985 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
986 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
987 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
988 	mtx_unlock(&ps->ps_mtx);
989 }
990 
991 /*
992  * kern_sigprocmask()
993  *
994  *	Manipulate signal mask.
995  */
996 int
997 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
998     int flags)
999 {
1000 	sigset_t new_block, oset1;
1001 	struct proc *p;
1002 	int error;
1003 
1004 	p = td->td_proc;
1005 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1006 		PROC_LOCK_ASSERT(p, MA_OWNED);
1007 	else
1008 		PROC_LOCK(p);
1009 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1010 	    ? MA_OWNED : MA_NOTOWNED);
1011 	if (oset != NULL)
1012 		*oset = td->td_sigmask;
1013 
1014 	error = 0;
1015 	if (set != NULL) {
1016 		switch (how) {
1017 		case SIG_BLOCK:
1018 			SIG_CANTMASK(*set);
1019 			oset1 = td->td_sigmask;
1020 			SIGSETOR(td->td_sigmask, *set);
1021 			new_block = td->td_sigmask;
1022 			SIGSETNAND(new_block, oset1);
1023 			break;
1024 		case SIG_UNBLOCK:
1025 			SIGSETNAND(td->td_sigmask, *set);
1026 			signotify(td);
1027 			goto out;
1028 		case SIG_SETMASK:
1029 			SIG_CANTMASK(*set);
1030 			oset1 = td->td_sigmask;
1031 			if (flags & SIGPROCMASK_OLD)
1032 				SIGSETLO(td->td_sigmask, *set);
1033 			else
1034 				td->td_sigmask = *set;
1035 			new_block = td->td_sigmask;
1036 			SIGSETNAND(new_block, oset1);
1037 			signotify(td);
1038 			break;
1039 		default:
1040 			error = EINVAL;
1041 			goto out;
1042 		}
1043 
1044 		/*
1045 		 * The new_block set contains signals that were not previously
1046 		 * blocked, but are blocked now.
1047 		 *
1048 		 * In case we block any signal that was not previously blocked
1049 		 * for td, and process has the signal pending, try to schedule
1050 		 * signal delivery to some thread that does not block the
1051 		 * signal, possibly waking it up.
1052 		 */
1053 		if (p->p_numthreads != 1)
1054 			reschedule_signals(p, new_block, flags);
1055 	}
1056 
1057 out:
1058 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1059 		PROC_UNLOCK(p);
1060 	return (error);
1061 }
1062 
1063 #ifndef _SYS_SYSPROTO_H_
1064 struct sigprocmask_args {
1065 	int	how;
1066 	const sigset_t *set;
1067 	sigset_t *oset;
1068 };
1069 #endif
1070 int
1071 sys_sigprocmask(td, uap)
1072 	register struct thread *td;
1073 	struct sigprocmask_args *uap;
1074 {
1075 	sigset_t set, oset;
1076 	sigset_t *setp, *osetp;
1077 	int error;
1078 
1079 	setp = (uap->set != NULL) ? &set : NULL;
1080 	osetp = (uap->oset != NULL) ? &oset : NULL;
1081 	if (setp) {
1082 		error = copyin(uap->set, setp, sizeof(set));
1083 		if (error)
1084 			return (error);
1085 	}
1086 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1087 	if (osetp && !error) {
1088 		error = copyout(osetp, uap->oset, sizeof(oset));
1089 	}
1090 	return (error);
1091 }
1092 
1093 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1094 #ifndef _SYS_SYSPROTO_H_
1095 struct osigprocmask_args {
1096 	int	how;
1097 	osigset_t mask;
1098 };
1099 #endif
1100 int
1101 osigprocmask(td, uap)
1102 	register struct thread *td;
1103 	struct osigprocmask_args *uap;
1104 {
1105 	sigset_t set, oset;
1106 	int error;
1107 
1108 	OSIG2SIG(uap->mask, set);
1109 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1110 	SIG2OSIG(oset, td->td_retval[0]);
1111 	return (error);
1112 }
1113 #endif /* COMPAT_43 */
1114 
1115 int
1116 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1117 {
1118 	ksiginfo_t ksi;
1119 	sigset_t set;
1120 	int error;
1121 
1122 	error = copyin(uap->set, &set, sizeof(set));
1123 	if (error) {
1124 		td->td_retval[0] = error;
1125 		return (0);
1126 	}
1127 
1128 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1129 	if (error) {
1130 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1131 			error = ERESTART;
1132 		if (error == ERESTART)
1133 			return (error);
1134 		td->td_retval[0] = error;
1135 		return (0);
1136 	}
1137 
1138 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1139 	td->td_retval[0] = error;
1140 	return (0);
1141 }
1142 
1143 int
1144 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1145 {
1146 	struct timespec ts;
1147 	struct timespec *timeout;
1148 	sigset_t set;
1149 	ksiginfo_t ksi;
1150 	int error;
1151 
1152 	if (uap->timeout) {
1153 		error = copyin(uap->timeout, &ts, sizeof(ts));
1154 		if (error)
1155 			return (error);
1156 
1157 		timeout = &ts;
1158 	} else
1159 		timeout = NULL;
1160 
1161 	error = copyin(uap->set, &set, sizeof(set));
1162 	if (error)
1163 		return (error);
1164 
1165 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1166 	if (error)
1167 		return (error);
1168 
1169 	if (uap->info)
1170 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1171 
1172 	if (error == 0)
1173 		td->td_retval[0] = ksi.ksi_signo;
1174 	return (error);
1175 }
1176 
1177 int
1178 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1179 {
1180 	ksiginfo_t ksi;
1181 	sigset_t set;
1182 	int error;
1183 
1184 	error = copyin(uap->set, &set, sizeof(set));
1185 	if (error)
1186 		return (error);
1187 
1188 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1189 	if (error)
1190 		return (error);
1191 
1192 	if (uap->info)
1193 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1194 
1195 	if (error == 0)
1196 		td->td_retval[0] = ksi.ksi_signo;
1197 	return (error);
1198 }
1199 
1200 int
1201 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1202 	struct timespec *timeout)
1203 {
1204 	struct sigacts *ps;
1205 	sigset_t saved_mask, new_block;
1206 	struct proc *p;
1207 	int error, sig, timo, timevalid = 0;
1208 	struct timespec rts, ets, ts;
1209 	struct timeval tv;
1210 
1211 	p = td->td_proc;
1212 	error = 0;
1213 	ets.tv_sec = 0;
1214 	ets.tv_nsec = 0;
1215 
1216 	if (timeout != NULL) {
1217 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1218 			timevalid = 1;
1219 			getnanouptime(&rts);
1220 			ets = rts;
1221 			timespecadd(&ets, timeout);
1222 		}
1223 	}
1224 	ksiginfo_init(ksi);
1225 	/* Some signals can not be waited for. */
1226 	SIG_CANTMASK(waitset);
1227 	ps = p->p_sigacts;
1228 	PROC_LOCK(p);
1229 	saved_mask = td->td_sigmask;
1230 	SIGSETNAND(td->td_sigmask, waitset);
1231 	for (;;) {
1232 		mtx_lock(&ps->ps_mtx);
1233 		sig = cursig(td);
1234 		mtx_unlock(&ps->ps_mtx);
1235 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1236 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1237 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1238 				error = 0;
1239 				break;
1240 			}
1241 		}
1242 
1243 		if (error != 0)
1244 			break;
1245 
1246 		/*
1247 		 * POSIX says this must be checked after looking for pending
1248 		 * signals.
1249 		 */
1250 		if (timeout != NULL) {
1251 			if (!timevalid) {
1252 				error = EINVAL;
1253 				break;
1254 			}
1255 			getnanouptime(&rts);
1256 			if (timespeccmp(&rts, &ets, >=)) {
1257 				error = EAGAIN;
1258 				break;
1259 			}
1260 			ts = ets;
1261 			timespecsub(&ts, &rts);
1262 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1263 			timo = tvtohz(&tv);
1264 		} else {
1265 			timo = 0;
1266 		}
1267 
1268 		error = msleep(ps, &p->p_mtx, PPAUSE|PCATCH, "sigwait", timo);
1269 
1270 		if (timeout != NULL) {
1271 			if (error == ERESTART) {
1272 				/* Timeout can not be restarted. */
1273 				error = EINTR;
1274 			} else if (error == EAGAIN) {
1275 				/* We will calculate timeout by ourself. */
1276 				error = 0;
1277 			}
1278 		}
1279 	}
1280 
1281 	new_block = saved_mask;
1282 	SIGSETNAND(new_block, td->td_sigmask);
1283 	td->td_sigmask = saved_mask;
1284 	/*
1285 	 * Fewer signals can be delivered to us, reschedule signal
1286 	 * notification.
1287 	 */
1288 	if (p->p_numthreads != 1)
1289 		reschedule_signals(p, new_block, 0);
1290 
1291 	if (error == 0) {
1292 		SDT_PROBE(proc, kernel, , signal__clear, sig, ksi, 0, 0, 0);
1293 
1294 		if (ksi->ksi_code == SI_TIMER)
1295 			itimer_accept(p, ksi->ksi_timerid, ksi);
1296 
1297 #ifdef KTRACE
1298 		if (KTRPOINT(td, KTR_PSIG)) {
1299 			sig_t action;
1300 
1301 			mtx_lock(&ps->ps_mtx);
1302 			action = ps->ps_sigact[_SIG_IDX(sig)];
1303 			mtx_unlock(&ps->ps_mtx);
1304 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1305 		}
1306 #endif
1307 		if (sig == SIGKILL)
1308 			sigexit(td, sig);
1309 	}
1310 	PROC_UNLOCK(p);
1311 	return (error);
1312 }
1313 
1314 #ifndef _SYS_SYSPROTO_H_
1315 struct sigpending_args {
1316 	sigset_t	*set;
1317 };
1318 #endif
1319 int
1320 sys_sigpending(td, uap)
1321 	struct thread *td;
1322 	struct sigpending_args *uap;
1323 {
1324 	struct proc *p = td->td_proc;
1325 	sigset_t pending;
1326 
1327 	PROC_LOCK(p);
1328 	pending = p->p_sigqueue.sq_signals;
1329 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1330 	PROC_UNLOCK(p);
1331 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1332 }
1333 
1334 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1335 #ifndef _SYS_SYSPROTO_H_
1336 struct osigpending_args {
1337 	int	dummy;
1338 };
1339 #endif
1340 int
1341 osigpending(td, uap)
1342 	struct thread *td;
1343 	struct osigpending_args *uap;
1344 {
1345 	struct proc *p = td->td_proc;
1346 	sigset_t pending;
1347 
1348 	PROC_LOCK(p);
1349 	pending = p->p_sigqueue.sq_signals;
1350 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1351 	PROC_UNLOCK(p);
1352 	SIG2OSIG(pending, td->td_retval[0]);
1353 	return (0);
1354 }
1355 #endif /* COMPAT_43 */
1356 
1357 #if defined(COMPAT_43)
1358 /*
1359  * Generalized interface signal handler, 4.3-compatible.
1360  */
1361 #ifndef _SYS_SYSPROTO_H_
1362 struct osigvec_args {
1363 	int	signum;
1364 	struct	sigvec *nsv;
1365 	struct	sigvec *osv;
1366 };
1367 #endif
1368 /* ARGSUSED */
1369 int
1370 osigvec(td, uap)
1371 	struct thread *td;
1372 	register struct osigvec_args *uap;
1373 {
1374 	struct sigvec vec;
1375 	struct sigaction nsa, osa;
1376 	register struct sigaction *nsap, *osap;
1377 	int error;
1378 
1379 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1380 		return (EINVAL);
1381 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1382 	osap = (uap->osv != NULL) ? &osa : NULL;
1383 	if (nsap) {
1384 		error = copyin(uap->nsv, &vec, sizeof(vec));
1385 		if (error)
1386 			return (error);
1387 		nsap->sa_handler = vec.sv_handler;
1388 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1389 		nsap->sa_flags = vec.sv_flags;
1390 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1391 	}
1392 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1393 	if (osap && !error) {
1394 		vec.sv_handler = osap->sa_handler;
1395 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1396 		vec.sv_flags = osap->sa_flags;
1397 		vec.sv_flags &= ~SA_NOCLDWAIT;
1398 		vec.sv_flags ^= SA_RESTART;
1399 		error = copyout(&vec, uap->osv, sizeof(vec));
1400 	}
1401 	return (error);
1402 }
1403 
1404 #ifndef _SYS_SYSPROTO_H_
1405 struct osigblock_args {
1406 	int	mask;
1407 };
1408 #endif
1409 int
1410 osigblock(td, uap)
1411 	register struct thread *td;
1412 	struct osigblock_args *uap;
1413 {
1414 	sigset_t set, oset;
1415 
1416 	OSIG2SIG(uap->mask, set);
1417 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1418 	SIG2OSIG(oset, td->td_retval[0]);
1419 	return (0);
1420 }
1421 
1422 #ifndef _SYS_SYSPROTO_H_
1423 struct osigsetmask_args {
1424 	int	mask;
1425 };
1426 #endif
1427 int
1428 osigsetmask(td, uap)
1429 	struct thread *td;
1430 	struct osigsetmask_args *uap;
1431 {
1432 	sigset_t set, oset;
1433 
1434 	OSIG2SIG(uap->mask, set);
1435 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1436 	SIG2OSIG(oset, td->td_retval[0]);
1437 	return (0);
1438 }
1439 #endif /* COMPAT_43 */
1440 
1441 /*
1442  * Suspend calling thread until signal, providing mask to be set in the
1443  * meantime.
1444  */
1445 #ifndef _SYS_SYSPROTO_H_
1446 struct sigsuspend_args {
1447 	const sigset_t *sigmask;
1448 };
1449 #endif
1450 /* ARGSUSED */
1451 int
1452 sys_sigsuspend(td, uap)
1453 	struct thread *td;
1454 	struct sigsuspend_args *uap;
1455 {
1456 	sigset_t mask;
1457 	int error;
1458 
1459 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1460 	if (error)
1461 		return (error);
1462 	return (kern_sigsuspend(td, mask));
1463 }
1464 
1465 int
1466 kern_sigsuspend(struct thread *td, sigset_t mask)
1467 {
1468 	struct proc *p = td->td_proc;
1469 	int has_sig, sig;
1470 
1471 	/*
1472 	 * When returning from sigsuspend, we want
1473 	 * the old mask to be restored after the
1474 	 * signal handler has finished.  Thus, we
1475 	 * save it here and mark the sigacts structure
1476 	 * to indicate this.
1477 	 */
1478 	PROC_LOCK(p);
1479 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1480 	    SIGPROCMASK_PROC_LOCKED);
1481 	td->td_pflags |= TDP_OLDMASK;
1482 
1483 	/*
1484 	 * Process signals now. Otherwise, we can get spurious wakeup
1485 	 * due to signal entered process queue, but delivered to other
1486 	 * thread. But sigsuspend should return only on signal
1487 	 * delivery.
1488 	 */
1489 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1490 	for (has_sig = 0; !has_sig;) {
1491 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1492 			0) == 0)
1493 			/* void */;
1494 		thread_suspend_check(0);
1495 		mtx_lock(&p->p_sigacts->ps_mtx);
1496 		while ((sig = cursig(td)) != 0)
1497 			has_sig += postsig(sig);
1498 		mtx_unlock(&p->p_sigacts->ps_mtx);
1499 	}
1500 	PROC_UNLOCK(p);
1501 	td->td_errno = EINTR;
1502 	td->td_pflags |= TDP_NERRNO;
1503 	return (EJUSTRETURN);
1504 }
1505 
1506 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1507 /*
1508  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1509  * convention: libc stub passes mask, not pointer, to save a copyin.
1510  */
1511 #ifndef _SYS_SYSPROTO_H_
1512 struct osigsuspend_args {
1513 	osigset_t mask;
1514 };
1515 #endif
1516 /* ARGSUSED */
1517 int
1518 osigsuspend(td, uap)
1519 	struct thread *td;
1520 	struct osigsuspend_args *uap;
1521 {
1522 	sigset_t mask;
1523 
1524 	OSIG2SIG(uap->mask, mask);
1525 	return (kern_sigsuspend(td, mask));
1526 }
1527 #endif /* COMPAT_43 */
1528 
1529 #if defined(COMPAT_43)
1530 #ifndef _SYS_SYSPROTO_H_
1531 struct osigstack_args {
1532 	struct	sigstack *nss;
1533 	struct	sigstack *oss;
1534 };
1535 #endif
1536 /* ARGSUSED */
1537 int
1538 osigstack(td, uap)
1539 	struct thread *td;
1540 	register struct osigstack_args *uap;
1541 {
1542 	struct sigstack nss, oss;
1543 	int error = 0;
1544 
1545 	if (uap->nss != NULL) {
1546 		error = copyin(uap->nss, &nss, sizeof(nss));
1547 		if (error)
1548 			return (error);
1549 	}
1550 	oss.ss_sp = td->td_sigstk.ss_sp;
1551 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1552 	if (uap->nss != NULL) {
1553 		td->td_sigstk.ss_sp = nss.ss_sp;
1554 		td->td_sigstk.ss_size = 0;
1555 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1556 		td->td_pflags |= TDP_ALTSTACK;
1557 	}
1558 	if (uap->oss != NULL)
1559 		error = copyout(&oss, uap->oss, sizeof(oss));
1560 
1561 	return (error);
1562 }
1563 #endif /* COMPAT_43 */
1564 
1565 #ifndef _SYS_SYSPROTO_H_
1566 struct sigaltstack_args {
1567 	stack_t	*ss;
1568 	stack_t	*oss;
1569 };
1570 #endif
1571 /* ARGSUSED */
1572 int
1573 sys_sigaltstack(td, uap)
1574 	struct thread *td;
1575 	register struct sigaltstack_args *uap;
1576 {
1577 	stack_t ss, oss;
1578 	int error;
1579 
1580 	if (uap->ss != NULL) {
1581 		error = copyin(uap->ss, &ss, sizeof(ss));
1582 		if (error)
1583 			return (error);
1584 	}
1585 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1586 	    (uap->oss != NULL) ? &oss : NULL);
1587 	if (error)
1588 		return (error);
1589 	if (uap->oss != NULL)
1590 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1591 	return (error);
1592 }
1593 
1594 int
1595 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1596 {
1597 	struct proc *p = td->td_proc;
1598 	int oonstack;
1599 
1600 	oonstack = sigonstack(cpu_getstack(td));
1601 
1602 	if (oss != NULL) {
1603 		*oss = td->td_sigstk;
1604 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1605 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1606 	}
1607 
1608 	if (ss != NULL) {
1609 		if (oonstack)
1610 			return (EPERM);
1611 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1612 			return (EINVAL);
1613 		if (!(ss->ss_flags & SS_DISABLE)) {
1614 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1615 				return (ENOMEM);
1616 
1617 			td->td_sigstk = *ss;
1618 			td->td_pflags |= TDP_ALTSTACK;
1619 		} else {
1620 			td->td_pflags &= ~TDP_ALTSTACK;
1621 		}
1622 	}
1623 	return (0);
1624 }
1625 
1626 /*
1627  * Common code for kill process group/broadcast kill.
1628  * cp is calling process.
1629  */
1630 static int
1631 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1632 {
1633 	struct proc *p;
1634 	struct pgrp *pgrp;
1635 	int err;
1636 	int ret;
1637 
1638 	ret = ESRCH;
1639 	if (all) {
1640 		/*
1641 		 * broadcast
1642 		 */
1643 		sx_slock(&allproc_lock);
1644 		FOREACH_PROC_IN_SYSTEM(p) {
1645 			PROC_LOCK(p);
1646 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1647 			    p == td->td_proc || p->p_state == PRS_NEW) {
1648 				PROC_UNLOCK(p);
1649 				continue;
1650 			}
1651 			err = p_cansignal(td, p, sig);
1652 			if (err == 0) {
1653 				if (sig)
1654 					pksignal(p, sig, ksi);
1655 				ret = err;
1656 			}
1657 			else if (ret == ESRCH)
1658 				ret = err;
1659 			PROC_UNLOCK(p);
1660 		}
1661 		sx_sunlock(&allproc_lock);
1662 	} else {
1663 		sx_slock(&proctree_lock);
1664 		if (pgid == 0) {
1665 			/*
1666 			 * zero pgid means send to my process group.
1667 			 */
1668 			pgrp = td->td_proc->p_pgrp;
1669 			PGRP_LOCK(pgrp);
1670 		} else {
1671 			pgrp = pgfind(pgid);
1672 			if (pgrp == NULL) {
1673 				sx_sunlock(&proctree_lock);
1674 				return (ESRCH);
1675 			}
1676 		}
1677 		sx_sunlock(&proctree_lock);
1678 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1679 			PROC_LOCK(p);
1680 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1681 			    p->p_state == PRS_NEW) {
1682 				PROC_UNLOCK(p);
1683 				continue;
1684 			}
1685 			err = p_cansignal(td, p, sig);
1686 			if (err == 0) {
1687 				if (sig)
1688 					pksignal(p, sig, ksi);
1689 				ret = err;
1690 			}
1691 			else if (ret == ESRCH)
1692 				ret = err;
1693 			PROC_UNLOCK(p);
1694 		}
1695 		PGRP_UNLOCK(pgrp);
1696 	}
1697 	return (ret);
1698 }
1699 
1700 #ifndef _SYS_SYSPROTO_H_
1701 struct kill_args {
1702 	int	pid;
1703 	int	signum;
1704 };
1705 #endif
1706 /* ARGSUSED */
1707 int
1708 sys_kill(struct thread *td, struct kill_args *uap)
1709 {
1710 	ksiginfo_t ksi;
1711 	struct proc *p;
1712 	int error;
1713 
1714 	/*
1715 	 * A process in capability mode can send signals only to himself.
1716 	 * The main rationale behind this is that abort(3) is implemented as
1717 	 * kill(getpid(), SIGABRT).
1718 	 */
1719 	if (IN_CAPABILITY_MODE(td) && uap->pid != td->td_proc->p_pid)
1720 		return (ECAPMODE);
1721 
1722 	AUDIT_ARG_SIGNUM(uap->signum);
1723 	AUDIT_ARG_PID(uap->pid);
1724 	if ((u_int)uap->signum > _SIG_MAXSIG)
1725 		return (EINVAL);
1726 
1727 	ksiginfo_init(&ksi);
1728 	ksi.ksi_signo = uap->signum;
1729 	ksi.ksi_code = SI_USER;
1730 	ksi.ksi_pid = td->td_proc->p_pid;
1731 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1732 
1733 	if (uap->pid > 0) {
1734 		/* kill single process */
1735 		if ((p = pfind(uap->pid)) == NULL) {
1736 			if ((p = zpfind(uap->pid)) == NULL)
1737 				return (ESRCH);
1738 		}
1739 		AUDIT_ARG_PROCESS(p);
1740 		error = p_cansignal(td, p, uap->signum);
1741 		if (error == 0 && uap->signum)
1742 			pksignal(p, uap->signum, &ksi);
1743 		PROC_UNLOCK(p);
1744 		return (error);
1745 	}
1746 	switch (uap->pid) {
1747 	case -1:		/* broadcast signal */
1748 		return (killpg1(td, uap->signum, 0, 1, &ksi));
1749 	case 0:			/* signal own process group */
1750 		return (killpg1(td, uap->signum, 0, 0, &ksi));
1751 	default:		/* negative explicit process group */
1752 		return (killpg1(td, uap->signum, -uap->pid, 0, &ksi));
1753 	}
1754 	/* NOTREACHED */
1755 }
1756 
1757 int
1758 sys_pdkill(td, uap)
1759 	struct thread *td;
1760 	struct pdkill_args *uap;
1761 {
1762 	struct proc *p;
1763 	cap_rights_t rights;
1764 	int error;
1765 
1766 	AUDIT_ARG_SIGNUM(uap->signum);
1767 	AUDIT_ARG_FD(uap->fd);
1768 	if ((u_int)uap->signum > _SIG_MAXSIG)
1769 		return (EINVAL);
1770 
1771 	error = procdesc_find(td, uap->fd,
1772 	    cap_rights_init(&rights, CAP_PDKILL), &p);
1773 	if (error)
1774 		return (error);
1775 	AUDIT_ARG_PROCESS(p);
1776 	error = p_cansignal(td, p, uap->signum);
1777 	if (error == 0 && uap->signum)
1778 		kern_psignal(p, uap->signum);
1779 	PROC_UNLOCK(p);
1780 	return (error);
1781 }
1782 
1783 #if defined(COMPAT_43)
1784 #ifndef _SYS_SYSPROTO_H_
1785 struct okillpg_args {
1786 	int	pgid;
1787 	int	signum;
1788 };
1789 #endif
1790 /* ARGSUSED */
1791 int
1792 okillpg(struct thread *td, struct okillpg_args *uap)
1793 {
1794 	ksiginfo_t ksi;
1795 
1796 	AUDIT_ARG_SIGNUM(uap->signum);
1797 	AUDIT_ARG_PID(uap->pgid);
1798 	if ((u_int)uap->signum > _SIG_MAXSIG)
1799 		return (EINVAL);
1800 
1801 	ksiginfo_init(&ksi);
1802 	ksi.ksi_signo = uap->signum;
1803 	ksi.ksi_code = SI_USER;
1804 	ksi.ksi_pid = td->td_proc->p_pid;
1805 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1806 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1807 }
1808 #endif /* COMPAT_43 */
1809 
1810 #ifndef _SYS_SYSPROTO_H_
1811 struct sigqueue_args {
1812 	pid_t pid;
1813 	int signum;
1814 	/* union sigval */ void *value;
1815 };
1816 #endif
1817 int
1818 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1819 {
1820 	ksiginfo_t ksi;
1821 	struct proc *p;
1822 	int error;
1823 
1824 	if ((u_int)uap->signum > _SIG_MAXSIG)
1825 		return (EINVAL);
1826 
1827 	/*
1828 	 * Specification says sigqueue can only send signal to
1829 	 * single process.
1830 	 */
1831 	if (uap->pid <= 0)
1832 		return (EINVAL);
1833 
1834 	if ((p = pfind(uap->pid)) == NULL) {
1835 		if ((p = zpfind(uap->pid)) == NULL)
1836 			return (ESRCH);
1837 	}
1838 	error = p_cansignal(td, p, uap->signum);
1839 	if (error == 0 && uap->signum != 0) {
1840 		ksiginfo_init(&ksi);
1841 		ksi.ksi_flags = KSI_SIGQ;
1842 		ksi.ksi_signo = uap->signum;
1843 		ksi.ksi_code = SI_QUEUE;
1844 		ksi.ksi_pid = td->td_proc->p_pid;
1845 		ksi.ksi_uid = td->td_ucred->cr_ruid;
1846 		ksi.ksi_value.sival_ptr = uap->value;
1847 		error = pksignal(p, ksi.ksi_signo, &ksi);
1848 	}
1849 	PROC_UNLOCK(p);
1850 	return (error);
1851 }
1852 
1853 /*
1854  * Send a signal to a process group.
1855  */
1856 void
1857 gsignal(int pgid, int sig, ksiginfo_t *ksi)
1858 {
1859 	struct pgrp *pgrp;
1860 
1861 	if (pgid != 0) {
1862 		sx_slock(&proctree_lock);
1863 		pgrp = pgfind(pgid);
1864 		sx_sunlock(&proctree_lock);
1865 		if (pgrp != NULL) {
1866 			pgsignal(pgrp, sig, 0, ksi);
1867 			PGRP_UNLOCK(pgrp);
1868 		}
1869 	}
1870 }
1871 
1872 /*
1873  * Send a signal to a process group.  If checktty is 1,
1874  * limit to members which have a controlling terminal.
1875  */
1876 void
1877 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
1878 {
1879 	struct proc *p;
1880 
1881 	if (pgrp) {
1882 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
1883 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1884 			PROC_LOCK(p);
1885 			if (p->p_state == PRS_NORMAL &&
1886 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
1887 				pksignal(p, sig, ksi);
1888 			PROC_UNLOCK(p);
1889 		}
1890 	}
1891 }
1892 
1893 
1894 /*
1895  * Recalculate the signal mask and reset the signal disposition after
1896  * usermode frame for delivery is formed.  Should be called after
1897  * mach-specific routine, because sysent->sv_sendsig() needs correct
1898  * ps_siginfo and signal mask.
1899  */
1900 static void
1901 postsig_done(int sig, struct thread *td, struct sigacts *ps)
1902 {
1903 	sigset_t mask;
1904 
1905 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1906 	td->td_ru.ru_nsignals++;
1907 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
1908 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
1909 		SIGADDSET(mask, sig);
1910 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
1911 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
1912 	if (SIGISMEMBER(ps->ps_sigreset, sig))
1913 		sigdflt(ps, sig);
1914 }
1915 
1916 
1917 /*
1918  * Send a signal caused by a trap to the current thread.  If it will be
1919  * caught immediately, deliver it with correct code.  Otherwise, post it
1920  * normally.
1921  */
1922 void
1923 trapsignal(struct thread *td, ksiginfo_t *ksi)
1924 {
1925 	struct sigacts *ps;
1926 	struct proc *p;
1927 	int sig;
1928 	int code;
1929 
1930 	p = td->td_proc;
1931 	sig = ksi->ksi_signo;
1932 	code = ksi->ksi_code;
1933 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
1934 
1935 	PROC_LOCK(p);
1936 	ps = p->p_sigacts;
1937 	mtx_lock(&ps->ps_mtx);
1938 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
1939 	    !SIGISMEMBER(td->td_sigmask, sig)) {
1940 #ifdef KTRACE
1941 		if (KTRPOINT(curthread, KTR_PSIG))
1942 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
1943 			    &td->td_sigmask, code);
1944 #endif
1945 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
1946 				ksi, &td->td_sigmask);
1947 		postsig_done(sig, td, ps);
1948 		mtx_unlock(&ps->ps_mtx);
1949 	} else {
1950 		/*
1951 		 * Avoid a possible infinite loop if the thread
1952 		 * masking the signal or process is ignoring the
1953 		 * signal.
1954 		 */
1955 		if (kern_forcesigexit &&
1956 		    (SIGISMEMBER(td->td_sigmask, sig) ||
1957 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
1958 			SIGDELSET(td->td_sigmask, sig);
1959 			SIGDELSET(ps->ps_sigcatch, sig);
1960 			SIGDELSET(ps->ps_sigignore, sig);
1961 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1962 		}
1963 		mtx_unlock(&ps->ps_mtx);
1964 		p->p_code = code;	/* XXX for core dump/debugger */
1965 		p->p_sig = sig;		/* XXX to verify code */
1966 		tdsendsignal(p, td, sig, ksi);
1967 	}
1968 	PROC_UNLOCK(p);
1969 }
1970 
1971 static struct thread *
1972 sigtd(struct proc *p, int sig, int prop)
1973 {
1974 	struct thread *td, *signal_td;
1975 
1976 	PROC_LOCK_ASSERT(p, MA_OWNED);
1977 
1978 	/*
1979 	 * Check if current thread can handle the signal without
1980 	 * switching context to another thread.
1981 	 */
1982 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig))
1983 		return (curthread);
1984 	signal_td = NULL;
1985 	FOREACH_THREAD_IN_PROC(p, td) {
1986 		if (!SIGISMEMBER(td->td_sigmask, sig)) {
1987 			signal_td = td;
1988 			break;
1989 		}
1990 	}
1991 	if (signal_td == NULL)
1992 		signal_td = FIRST_THREAD_IN_PROC(p);
1993 	return (signal_td);
1994 }
1995 
1996 /*
1997  * Send the signal to the process.  If the signal has an action, the action
1998  * is usually performed by the target process rather than the caller; we add
1999  * the signal to the set of pending signals for the process.
2000  *
2001  * Exceptions:
2002  *   o When a stop signal is sent to a sleeping process that takes the
2003  *     default action, the process is stopped without awakening it.
2004  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2005  *     regardless of the signal action (eg, blocked or ignored).
2006  *
2007  * Other ignored signals are discarded immediately.
2008  *
2009  * NB: This function may be entered from the debugger via the "kill" DDB
2010  * command.  There is little that can be done to mitigate the possibly messy
2011  * side effects of this unwise possibility.
2012  */
2013 void
2014 kern_psignal(struct proc *p, int sig)
2015 {
2016 	ksiginfo_t ksi;
2017 
2018 	ksiginfo_init(&ksi);
2019 	ksi.ksi_signo = sig;
2020 	ksi.ksi_code = SI_KERNEL;
2021 	(void) tdsendsignal(p, NULL, sig, &ksi);
2022 }
2023 
2024 int
2025 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2026 {
2027 
2028 	return (tdsendsignal(p, NULL, sig, ksi));
2029 }
2030 
2031 /* Utility function for finding a thread to send signal event to. */
2032 int
2033 sigev_findtd(struct proc *p ,struct sigevent *sigev, struct thread **ttd)
2034 {
2035 	struct thread *td;
2036 
2037 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2038 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2039 		if (td == NULL)
2040 			return (ESRCH);
2041 		*ttd = td;
2042 	} else {
2043 		*ttd = NULL;
2044 		PROC_LOCK(p);
2045 	}
2046 	return (0);
2047 }
2048 
2049 void
2050 tdsignal(struct thread *td, int sig)
2051 {
2052 	ksiginfo_t ksi;
2053 
2054 	ksiginfo_init(&ksi);
2055 	ksi.ksi_signo = sig;
2056 	ksi.ksi_code = SI_KERNEL;
2057 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2058 }
2059 
2060 void
2061 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2062 {
2063 
2064 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2065 }
2066 
2067 int
2068 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2069 {
2070 	sig_t action;
2071 	sigqueue_t *sigqueue;
2072 	int prop;
2073 	struct sigacts *ps;
2074 	int intrval;
2075 	int ret = 0;
2076 	int wakeup_swapper;
2077 
2078 	MPASS(td == NULL || p == td->td_proc);
2079 	PROC_LOCK_ASSERT(p, MA_OWNED);
2080 
2081 	if (!_SIG_VALID(sig))
2082 		panic("%s(): invalid signal %d", __func__, sig);
2083 
2084 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2085 
2086 	/*
2087 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2088 	 */
2089 	if (p->p_state == PRS_ZOMBIE) {
2090 		if (ksi && (ksi->ksi_flags & KSI_INS))
2091 			ksiginfo_tryfree(ksi);
2092 		return (ret);
2093 	}
2094 
2095 	ps = p->p_sigacts;
2096 	KNOTE_LOCKED(&p->p_klist, NOTE_SIGNAL | sig);
2097 	prop = sigprop(sig);
2098 
2099 	if (td == NULL) {
2100 		td = sigtd(p, sig, prop);
2101 		sigqueue = &p->p_sigqueue;
2102 	} else
2103 		sigqueue = &td->td_sigqueue;
2104 
2105 	SDT_PROBE(proc, kernel, , signal__send, td, p, sig, 0, 0 );
2106 
2107 	/*
2108 	 * If the signal is being ignored,
2109 	 * then we forget about it immediately.
2110 	 * (Note: we don't set SIGCONT in ps_sigignore,
2111 	 * and if it is set to SIG_IGN,
2112 	 * action will be SIG_DFL here.)
2113 	 */
2114 	mtx_lock(&ps->ps_mtx);
2115 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2116 		SDT_PROBE(proc, kernel, , signal__discard, td, p, sig, 0, 0 );
2117 
2118 		mtx_unlock(&ps->ps_mtx);
2119 		if (ksi && (ksi->ksi_flags & KSI_INS))
2120 			ksiginfo_tryfree(ksi);
2121 		return (ret);
2122 	}
2123 	if (SIGISMEMBER(td->td_sigmask, sig))
2124 		action = SIG_HOLD;
2125 	else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2126 		action = SIG_CATCH;
2127 	else
2128 		action = SIG_DFL;
2129 	if (SIGISMEMBER(ps->ps_sigintr, sig))
2130 		intrval = EINTR;
2131 	else
2132 		intrval = ERESTART;
2133 	mtx_unlock(&ps->ps_mtx);
2134 
2135 	if (prop & SA_CONT)
2136 		sigqueue_delete_stopmask_proc(p);
2137 	else if (prop & SA_STOP) {
2138 		/*
2139 		 * If sending a tty stop signal to a member of an orphaned
2140 		 * process group, discard the signal here if the action
2141 		 * is default; don't stop the process below if sleeping,
2142 		 * and don't clear any pending SIGCONT.
2143 		 */
2144 		if ((prop & SA_TTYSTOP) &&
2145 		    (p->p_pgrp->pg_jobc == 0) &&
2146 		    (action == SIG_DFL)) {
2147 			if (ksi && (ksi->ksi_flags & KSI_INS))
2148 				ksiginfo_tryfree(ksi);
2149 			return (ret);
2150 		}
2151 		sigqueue_delete_proc(p, SIGCONT);
2152 		if (p->p_flag & P_CONTINUED) {
2153 			p->p_flag &= ~P_CONTINUED;
2154 			PROC_LOCK(p->p_pptr);
2155 			sigqueue_take(p->p_ksi);
2156 			PROC_UNLOCK(p->p_pptr);
2157 		}
2158 	}
2159 
2160 	ret = sigqueue_add(sigqueue, sig, ksi);
2161 	if (ret != 0)
2162 		return (ret);
2163 	signotify(td);
2164 	/*
2165 	 * Defer further processing for signals which are held,
2166 	 * except that stopped processes must be continued by SIGCONT.
2167 	 */
2168 	if (action == SIG_HOLD &&
2169 	    !((prop & SA_CONT) && (p->p_flag & P_STOPPED_SIG)))
2170 		return (ret);
2171 	/*
2172 	 * SIGKILL: Remove procfs STOPEVENTs.
2173 	 */
2174 	if (sig == SIGKILL) {
2175 		/* from procfs_ioctl.c: PIOCBIC */
2176 		p->p_stops = 0;
2177 		/* from procfs_ioctl.c: PIOCCONT */
2178 		p->p_step = 0;
2179 		wakeup(&p->p_step);
2180 	}
2181 	/*
2182 	 * Some signals have a process-wide effect and a per-thread
2183 	 * component.  Most processing occurs when the process next
2184 	 * tries to cross the user boundary, however there are some
2185 	 * times when processing needs to be done immediately, such as
2186 	 * waking up threads so that they can cross the user boundary.
2187 	 * We try to do the per-process part here.
2188 	 */
2189 	if (P_SHOULDSTOP(p)) {
2190 		KASSERT(!(p->p_flag & P_WEXIT),
2191 		    ("signal to stopped but exiting process"));
2192 		if (sig == SIGKILL) {
2193 			/*
2194 			 * If traced process is already stopped,
2195 			 * then no further action is necessary.
2196 			 */
2197 			if (p->p_flag & P_TRACED)
2198 				goto out;
2199 			/*
2200 			 * SIGKILL sets process running.
2201 			 * It will die elsewhere.
2202 			 * All threads must be restarted.
2203 			 */
2204 			p->p_flag &= ~P_STOPPED_SIG;
2205 			goto runfast;
2206 		}
2207 
2208 		if (prop & SA_CONT) {
2209 			/*
2210 			 * If traced process is already stopped,
2211 			 * then no further action is necessary.
2212 			 */
2213 			if (p->p_flag & P_TRACED)
2214 				goto out;
2215 			/*
2216 			 * If SIGCONT is default (or ignored), we continue the
2217 			 * process but don't leave the signal in sigqueue as
2218 			 * it has no further action.  If SIGCONT is held, we
2219 			 * continue the process and leave the signal in
2220 			 * sigqueue.  If the process catches SIGCONT, let it
2221 			 * handle the signal itself.  If it isn't waiting on
2222 			 * an event, it goes back to run state.
2223 			 * Otherwise, process goes back to sleep state.
2224 			 */
2225 			p->p_flag &= ~P_STOPPED_SIG;
2226 			PROC_SLOCK(p);
2227 			if (p->p_numthreads == p->p_suspcount) {
2228 				PROC_SUNLOCK(p);
2229 				p->p_flag |= P_CONTINUED;
2230 				p->p_xsig = SIGCONT;
2231 				PROC_LOCK(p->p_pptr);
2232 				childproc_continued(p);
2233 				PROC_UNLOCK(p->p_pptr);
2234 				PROC_SLOCK(p);
2235 			}
2236 			if (action == SIG_DFL) {
2237 				thread_unsuspend(p);
2238 				PROC_SUNLOCK(p);
2239 				sigqueue_delete(sigqueue, sig);
2240 				goto out;
2241 			}
2242 			if (action == SIG_CATCH) {
2243 				/*
2244 				 * The process wants to catch it so it needs
2245 				 * to run at least one thread, but which one?
2246 				 */
2247 				PROC_SUNLOCK(p);
2248 				goto runfast;
2249 			}
2250 			/*
2251 			 * The signal is not ignored or caught.
2252 			 */
2253 			thread_unsuspend(p);
2254 			PROC_SUNLOCK(p);
2255 			goto out;
2256 		}
2257 
2258 		if (prop & SA_STOP) {
2259 			/*
2260 			 * If traced process is already stopped,
2261 			 * then no further action is necessary.
2262 			 */
2263 			if (p->p_flag & P_TRACED)
2264 				goto out;
2265 			/*
2266 			 * Already stopped, don't need to stop again
2267 			 * (If we did the shell could get confused).
2268 			 * Just make sure the signal STOP bit set.
2269 			 */
2270 			p->p_flag |= P_STOPPED_SIG;
2271 			sigqueue_delete(sigqueue, sig);
2272 			goto out;
2273 		}
2274 
2275 		/*
2276 		 * All other kinds of signals:
2277 		 * If a thread is sleeping interruptibly, simulate a
2278 		 * wakeup so that when it is continued it will be made
2279 		 * runnable and can look at the signal.  However, don't make
2280 		 * the PROCESS runnable, leave it stopped.
2281 		 * It may run a bit until it hits a thread_suspend_check().
2282 		 */
2283 		wakeup_swapper = 0;
2284 		PROC_SLOCK(p);
2285 		thread_lock(td);
2286 		if (TD_ON_SLEEPQ(td) && (td->td_flags & TDF_SINTR))
2287 			wakeup_swapper = sleepq_abort(td, intrval);
2288 		thread_unlock(td);
2289 		PROC_SUNLOCK(p);
2290 		if (wakeup_swapper)
2291 			kick_proc0();
2292 		goto out;
2293 		/*
2294 		 * Mutexes are short lived. Threads waiting on them will
2295 		 * hit thread_suspend_check() soon.
2296 		 */
2297 	} else if (p->p_state == PRS_NORMAL) {
2298 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2299 			tdsigwakeup(td, sig, action, intrval);
2300 			goto out;
2301 		}
2302 
2303 		MPASS(action == SIG_DFL);
2304 
2305 		if (prop & SA_STOP) {
2306 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2307 				goto out;
2308 			p->p_flag |= P_STOPPED_SIG;
2309 			p->p_xsig = sig;
2310 			PROC_SLOCK(p);
2311 			sig_suspend_threads(td, p, 1);
2312 			if (p->p_numthreads == p->p_suspcount) {
2313 				/*
2314 				 * only thread sending signal to another
2315 				 * process can reach here, if thread is sending
2316 				 * signal to its process, because thread does
2317 				 * not suspend itself here, p_numthreads
2318 				 * should never be equal to p_suspcount.
2319 				 */
2320 				thread_stopped(p);
2321 				PROC_SUNLOCK(p);
2322 				sigqueue_delete_proc(p, p->p_xsig);
2323 			} else
2324 				PROC_SUNLOCK(p);
2325 			goto out;
2326 		}
2327 	} else {
2328 		/* Not in "NORMAL" state. discard the signal. */
2329 		sigqueue_delete(sigqueue, sig);
2330 		goto out;
2331 	}
2332 
2333 	/*
2334 	 * The process is not stopped so we need to apply the signal to all the
2335 	 * running threads.
2336 	 */
2337 runfast:
2338 	tdsigwakeup(td, sig, action, intrval);
2339 	PROC_SLOCK(p);
2340 	thread_unsuspend(p);
2341 	PROC_SUNLOCK(p);
2342 out:
2343 	/* If we jump here, proc slock should not be owned. */
2344 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2345 	return (ret);
2346 }
2347 
2348 /*
2349  * The force of a signal has been directed against a single
2350  * thread.  We need to see what we can do about knocking it
2351  * out of any sleep it may be in etc.
2352  */
2353 static void
2354 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2355 {
2356 	struct proc *p = td->td_proc;
2357 	register int prop;
2358 	int wakeup_swapper;
2359 
2360 	wakeup_swapper = 0;
2361 	PROC_LOCK_ASSERT(p, MA_OWNED);
2362 	prop = sigprop(sig);
2363 
2364 	PROC_SLOCK(p);
2365 	thread_lock(td);
2366 	/*
2367 	 * Bring the priority of a thread up if we want it to get
2368 	 * killed in this lifetime.  Be careful to avoid bumping the
2369 	 * priority of the idle thread, since we still allow to signal
2370 	 * kernel processes.
2371 	 */
2372 	if (action == SIG_DFL && (prop & SA_KILL) != 0 &&
2373 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2374 		sched_prio(td, PUSER);
2375 	if (TD_ON_SLEEPQ(td)) {
2376 		/*
2377 		 * If thread is sleeping uninterruptibly
2378 		 * we can't interrupt the sleep... the signal will
2379 		 * be noticed when the process returns through
2380 		 * trap() or syscall().
2381 		 */
2382 		if ((td->td_flags & TDF_SINTR) == 0)
2383 			goto out;
2384 		/*
2385 		 * If SIGCONT is default (or ignored) and process is
2386 		 * asleep, we are finished; the process should not
2387 		 * be awakened.
2388 		 */
2389 		if ((prop & SA_CONT) && action == SIG_DFL) {
2390 			thread_unlock(td);
2391 			PROC_SUNLOCK(p);
2392 			sigqueue_delete(&p->p_sigqueue, sig);
2393 			/*
2394 			 * It may be on either list in this state.
2395 			 * Remove from both for now.
2396 			 */
2397 			sigqueue_delete(&td->td_sigqueue, sig);
2398 			return;
2399 		}
2400 
2401 		/*
2402 		 * Don't awaken a sleeping thread for SIGSTOP if the
2403 		 * STOP signal is deferred.
2404 		 */
2405 		if ((prop & SA_STOP) && (td->td_flags & TDF_SBDRY))
2406 			goto out;
2407 
2408 		/*
2409 		 * Give low priority threads a better chance to run.
2410 		 */
2411 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2412 			sched_prio(td, PUSER);
2413 
2414 		wakeup_swapper = sleepq_abort(td, intrval);
2415 	} else {
2416 		/*
2417 		 * Other states do nothing with the signal immediately,
2418 		 * other than kicking ourselves if we are running.
2419 		 * It will either never be noticed, or noticed very soon.
2420 		 */
2421 #ifdef SMP
2422 		if (TD_IS_RUNNING(td) && td != curthread)
2423 			forward_signal(td);
2424 #endif
2425 	}
2426 out:
2427 	PROC_SUNLOCK(p);
2428 	thread_unlock(td);
2429 	if (wakeup_swapper)
2430 		kick_proc0();
2431 }
2432 
2433 static void
2434 sig_suspend_threads(struct thread *td, struct proc *p, int sending)
2435 {
2436 	struct thread *td2;
2437 
2438 	PROC_LOCK_ASSERT(p, MA_OWNED);
2439 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2440 
2441 	FOREACH_THREAD_IN_PROC(p, td2) {
2442 		thread_lock(td2);
2443 		td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
2444 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2445 		    (td2->td_flags & TDF_SINTR)) {
2446 			if (td2->td_flags & TDF_SBDRY) {
2447 				/*
2448 				 * Once a thread is asleep with
2449 				 * TDF_SBDRY set, it should never
2450 				 * become suspended due to this check.
2451 				 */
2452 				KASSERT(!TD_IS_SUSPENDED(td2),
2453 				    ("thread with deferred stops suspended"));
2454 			} else if (!TD_IS_SUSPENDED(td2)) {
2455 				thread_suspend_one(td2);
2456 			}
2457 		} else if (!TD_IS_SUSPENDED(td2)) {
2458 			if (sending || td != td2)
2459 				td2->td_flags |= TDF_ASTPENDING;
2460 #ifdef SMP
2461 			if (TD_IS_RUNNING(td2) && td2 != td)
2462 				forward_signal(td2);
2463 #endif
2464 		}
2465 		thread_unlock(td2);
2466 	}
2467 }
2468 
2469 int
2470 ptracestop(struct thread *td, int sig)
2471 {
2472 	struct proc *p = td->td_proc;
2473 
2474 	PROC_LOCK_ASSERT(p, MA_OWNED);
2475 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2476 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2477 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2478 
2479 	td->td_dbgflags |= TDB_XSIG;
2480 	td->td_xsig = sig;
2481 	CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2482 	    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2483 	PROC_SLOCK(p);
2484 	while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2485 		if (p->p_flag & P_SINGLE_EXIT) {
2486 			td->td_dbgflags &= ~TDB_XSIG;
2487 			PROC_SUNLOCK(p);
2488 			return (sig);
2489 		}
2490 		/*
2491 		 * Just make wait() to work, the last stopped thread
2492 		 * will win.
2493 		 */
2494 		p->p_xsig = sig;
2495 		p->p_xthread = td;
2496 		p->p_flag |= (P_STOPPED_SIG|P_STOPPED_TRACE);
2497 		sig_suspend_threads(td, p, 0);
2498 		if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2499 			td->td_dbgflags &= ~TDB_STOPATFORK;
2500 			cv_broadcast(&p->p_dbgwait);
2501 		}
2502 stopme:
2503 		thread_suspend_switch(td, p);
2504 		if (p->p_xthread == td)
2505 			p->p_xthread = NULL;
2506 		if (!(p->p_flag & P_TRACED))
2507 			break;
2508 		if (td->td_dbgflags & TDB_SUSPEND) {
2509 			if (p->p_flag & P_SINGLE_EXIT)
2510 				break;
2511 			goto stopme;
2512 		}
2513 	}
2514 	PROC_SUNLOCK(p);
2515 	return (td->td_xsig);
2516 }
2517 
2518 static void
2519 reschedule_signals(struct proc *p, sigset_t block, int flags)
2520 {
2521 	struct sigacts *ps;
2522 	struct thread *td;
2523 	int sig;
2524 
2525 	PROC_LOCK_ASSERT(p, MA_OWNED);
2526 	ps = p->p_sigacts;
2527 	mtx_assert(&ps->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0 ?
2528 	    MA_OWNED : MA_NOTOWNED);
2529 	if (SIGISEMPTY(p->p_siglist))
2530 		return;
2531 	SIGSETAND(block, p->p_siglist);
2532 	while ((sig = sig_ffs(&block)) != 0) {
2533 		SIGDELSET(block, sig);
2534 		td = sigtd(p, sig, 0);
2535 		signotify(td);
2536 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2537 			mtx_lock(&ps->ps_mtx);
2538 		if (p->p_flag & P_TRACED || SIGISMEMBER(ps->ps_sigcatch, sig))
2539 			tdsigwakeup(td, sig, SIG_CATCH,
2540 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2541 			     ERESTART));
2542 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2543 			mtx_unlock(&ps->ps_mtx);
2544 	}
2545 }
2546 
2547 void
2548 tdsigcleanup(struct thread *td)
2549 {
2550 	struct proc *p;
2551 	sigset_t unblocked;
2552 
2553 	p = td->td_proc;
2554 	PROC_LOCK_ASSERT(p, MA_OWNED);
2555 
2556 	sigqueue_flush(&td->td_sigqueue);
2557 	if (p->p_numthreads == 1)
2558 		return;
2559 
2560 	/*
2561 	 * Since we cannot handle signals, notify signal post code
2562 	 * about this by filling the sigmask.
2563 	 *
2564 	 * Also, if needed, wake up thread(s) that do not block the
2565 	 * same signals as the exiting thread, since the thread might
2566 	 * have been selected for delivery and woken up.
2567 	 */
2568 	SIGFILLSET(unblocked);
2569 	SIGSETNAND(unblocked, td->td_sigmask);
2570 	SIGFILLSET(td->td_sigmask);
2571 	reschedule_signals(p, unblocked, 0);
2572 
2573 }
2574 
2575 /*
2576  * Defer the delivery of SIGSTOP for the current thread.  Returns true
2577  * if stops were deferred and false if they were already deferred.
2578  */
2579 int
2580 sigdeferstop(void)
2581 {
2582 	struct thread *td;
2583 
2584 	td = curthread;
2585 	if (td->td_flags & TDF_SBDRY)
2586 		return (0);
2587 	thread_lock(td);
2588 	td->td_flags |= TDF_SBDRY;
2589 	thread_unlock(td);
2590 	return (1);
2591 }
2592 
2593 /*
2594  * Permit the delivery of SIGSTOP for the current thread.  This does
2595  * not immediately suspend if a stop was posted.  Instead, the thread
2596  * will suspend either via ast() or a subsequent interruptible sleep.
2597  */
2598 int
2599 sigallowstop(void)
2600 {
2601 	struct thread *td;
2602 	int prev;
2603 
2604 	td = curthread;
2605 	thread_lock(td);
2606 	prev = (td->td_flags & TDF_SBDRY) != 0;
2607 	td->td_flags &= ~TDF_SBDRY;
2608 	thread_unlock(td);
2609 	return (prev);
2610 }
2611 
2612 /*
2613  * If the current process has received a signal (should be caught or cause
2614  * termination, should interrupt current syscall), return the signal number.
2615  * Stop signals with default action are processed immediately, then cleared;
2616  * they aren't returned.  This is checked after each entry to the system for
2617  * a syscall or trap (though this can usually be done without calling issignal
2618  * by checking the pending signal masks in cursig.) The normal call
2619  * sequence is
2620  *
2621  *	while (sig = cursig(curthread))
2622  *		postsig(sig);
2623  */
2624 static int
2625 issignal(struct thread *td)
2626 {
2627 	struct proc *p;
2628 	struct sigacts *ps;
2629 	struct sigqueue *queue;
2630 	sigset_t sigpending;
2631 	int sig, prop, newsig;
2632 
2633 	p = td->td_proc;
2634 	ps = p->p_sigacts;
2635 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2636 	PROC_LOCK_ASSERT(p, MA_OWNED);
2637 	for (;;) {
2638 		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
2639 
2640 		sigpending = td->td_sigqueue.sq_signals;
2641 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
2642 		SIGSETNAND(sigpending, td->td_sigmask);
2643 
2644 		if (p->p_flag & P_PPWAIT || td->td_flags & TDF_SBDRY)
2645 			SIG_STOPSIGMASK(sigpending);
2646 		if (SIGISEMPTY(sigpending))	/* no signal to send */
2647 			return (0);
2648 		sig = sig_ffs(&sigpending);
2649 
2650 		if (p->p_stops & S_SIG) {
2651 			mtx_unlock(&ps->ps_mtx);
2652 			stopevent(p, S_SIG, sig);
2653 			mtx_lock(&ps->ps_mtx);
2654 		}
2655 
2656 		/*
2657 		 * We should see pending but ignored signals
2658 		 * only if P_TRACED was on when they were posted.
2659 		 */
2660 		if (SIGISMEMBER(ps->ps_sigignore, sig) && (traced == 0)) {
2661 			sigqueue_delete(&td->td_sigqueue, sig);
2662 			sigqueue_delete(&p->p_sigqueue, sig);
2663 			continue;
2664 		}
2665 		if (p->p_flag & P_TRACED && (p->p_flag & P_PPTRACE) == 0) {
2666 			/*
2667 			 * If traced, always stop.
2668 			 * Remove old signal from queue before the stop.
2669 			 * XXX shrug off debugger, it causes siginfo to
2670 			 * be thrown away.
2671 			 */
2672 			queue = &td->td_sigqueue;
2673 			td->td_dbgksi.ksi_signo = 0;
2674 			if (sigqueue_get(queue, sig, &td->td_dbgksi) == 0) {
2675 				queue = &p->p_sigqueue;
2676 				sigqueue_get(queue, sig, &td->td_dbgksi);
2677 			}
2678 
2679 			mtx_unlock(&ps->ps_mtx);
2680 			newsig = ptracestop(td, sig);
2681 			mtx_lock(&ps->ps_mtx);
2682 
2683 			if (sig != newsig) {
2684 
2685 				/*
2686 				 * If parent wants us to take the signal,
2687 				 * then it will leave it in p->p_xsig;
2688 				 * otherwise we just look for signals again.
2689 				*/
2690 				if (newsig == 0)
2691 					continue;
2692 				sig = newsig;
2693 
2694 				/*
2695 				 * Put the new signal into td_sigqueue. If the
2696 				 * signal is being masked, look for other
2697 				 * signals.
2698 				 */
2699 				sigqueue_add(queue, sig, NULL);
2700 				if (SIGISMEMBER(td->td_sigmask, sig))
2701 					continue;
2702 				signotify(td);
2703 			} else {
2704 				if (td->td_dbgksi.ksi_signo != 0) {
2705 					td->td_dbgksi.ksi_flags |= KSI_HEAD;
2706 					if (sigqueue_add(&td->td_sigqueue, sig,
2707 					    &td->td_dbgksi) != 0)
2708 						td->td_dbgksi.ksi_signo = 0;
2709 				}
2710 				if (td->td_dbgksi.ksi_signo == 0)
2711 					sigqueue_add(&td->td_sigqueue, sig,
2712 					    NULL);
2713 			}
2714 
2715 			/*
2716 			 * If the traced bit got turned off, go back up
2717 			 * to the top to rescan signals.  This ensures
2718 			 * that p_sig* and p_sigact are consistent.
2719 			 */
2720 			if ((p->p_flag & P_TRACED) == 0)
2721 				continue;
2722 		}
2723 
2724 		prop = sigprop(sig);
2725 
2726 		/*
2727 		 * Decide whether the signal should be returned.
2728 		 * Return the signal's number, or fall through
2729 		 * to clear it from the pending mask.
2730 		 */
2731 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2732 
2733 		case (intptr_t)SIG_DFL:
2734 			/*
2735 			 * Don't take default actions on system processes.
2736 			 */
2737 			if (p->p_pid <= 1) {
2738 #ifdef DIAGNOSTIC
2739 				/*
2740 				 * Are you sure you want to ignore SIGSEGV
2741 				 * in init? XXX
2742 				 */
2743 				printf("Process (pid %lu) got signal %d\n",
2744 					(u_long)p->p_pid, sig);
2745 #endif
2746 				break;		/* == ignore */
2747 			}
2748 			/*
2749 			 * If there is a pending stop signal to process
2750 			 * with default action, stop here,
2751 			 * then clear the signal.  However,
2752 			 * if process is member of an orphaned
2753 			 * process group, ignore tty stop signals.
2754 			 */
2755 			if (prop & SA_STOP) {
2756 				if (p->p_flag & (P_TRACED|P_WEXIT) ||
2757 				    (p->p_pgrp->pg_jobc == 0 &&
2758 				     prop & SA_TTYSTOP))
2759 					break;	/* == ignore */
2760 				mtx_unlock(&ps->ps_mtx);
2761 				WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2762 				    &p->p_mtx.lock_object, "Catching SIGSTOP");
2763 				p->p_flag |= P_STOPPED_SIG;
2764 				p->p_xsig = sig;
2765 				PROC_SLOCK(p);
2766 				sig_suspend_threads(td, p, 0);
2767 				thread_suspend_switch(td, p);
2768 				PROC_SUNLOCK(p);
2769 				mtx_lock(&ps->ps_mtx);
2770 				break;
2771 			} else if (prop & SA_IGNORE) {
2772 				/*
2773 				 * Except for SIGCONT, shouldn't get here.
2774 				 * Default action is to ignore; drop it.
2775 				 */
2776 				break;		/* == ignore */
2777 			} else
2778 				return (sig);
2779 			/*NOTREACHED*/
2780 
2781 		case (intptr_t)SIG_IGN:
2782 			/*
2783 			 * Masking above should prevent us ever trying
2784 			 * to take action on an ignored signal other
2785 			 * than SIGCONT, unless process is traced.
2786 			 */
2787 			if ((prop & SA_CONT) == 0 &&
2788 			    (p->p_flag & P_TRACED) == 0)
2789 				printf("issignal\n");
2790 			break;		/* == ignore */
2791 
2792 		default:
2793 			/*
2794 			 * This signal has an action, let
2795 			 * postsig() process it.
2796 			 */
2797 			return (sig);
2798 		}
2799 		sigqueue_delete(&td->td_sigqueue, sig);	/* take the signal! */
2800 		sigqueue_delete(&p->p_sigqueue, sig);
2801 	}
2802 	/* NOTREACHED */
2803 }
2804 
2805 void
2806 thread_stopped(struct proc *p)
2807 {
2808 	int n;
2809 
2810 	PROC_LOCK_ASSERT(p, MA_OWNED);
2811 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2812 	n = p->p_suspcount;
2813 	if (p == curproc)
2814 		n++;
2815 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
2816 		PROC_SUNLOCK(p);
2817 		p->p_flag &= ~P_WAITED;
2818 		PROC_LOCK(p->p_pptr);
2819 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
2820 			CLD_TRAPPED : CLD_STOPPED);
2821 		PROC_UNLOCK(p->p_pptr);
2822 		PROC_SLOCK(p);
2823 	}
2824 }
2825 
2826 /*
2827  * Take the action for the specified signal
2828  * from the current set of pending signals.
2829  */
2830 int
2831 postsig(sig)
2832 	register int sig;
2833 {
2834 	struct thread *td = curthread;
2835 	register struct proc *p = td->td_proc;
2836 	struct sigacts *ps;
2837 	sig_t action;
2838 	ksiginfo_t ksi;
2839 	sigset_t returnmask;
2840 
2841 	KASSERT(sig != 0, ("postsig"));
2842 
2843 	PROC_LOCK_ASSERT(p, MA_OWNED);
2844 	ps = p->p_sigacts;
2845 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2846 	ksiginfo_init(&ksi);
2847 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
2848 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
2849 		return (0);
2850 	ksi.ksi_signo = sig;
2851 	if (ksi.ksi_code == SI_TIMER)
2852 		itimer_accept(p, ksi.ksi_timerid, &ksi);
2853 	action = ps->ps_sigact[_SIG_IDX(sig)];
2854 #ifdef KTRACE
2855 	if (KTRPOINT(td, KTR_PSIG))
2856 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
2857 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
2858 #endif
2859 	if (p->p_stops & S_SIG) {
2860 		mtx_unlock(&ps->ps_mtx);
2861 		stopevent(p, S_SIG, sig);
2862 		mtx_lock(&ps->ps_mtx);
2863 	}
2864 
2865 	if (action == SIG_DFL) {
2866 		/*
2867 		 * Default action, where the default is to kill
2868 		 * the process.  (Other cases were ignored above.)
2869 		 */
2870 		mtx_unlock(&ps->ps_mtx);
2871 		sigexit(td, sig);
2872 		/* NOTREACHED */
2873 	} else {
2874 		/*
2875 		 * If we get here, the signal must be caught.
2876 		 */
2877 		KASSERT(action != SIG_IGN && !SIGISMEMBER(td->td_sigmask, sig),
2878 		    ("postsig action"));
2879 		/*
2880 		 * Set the new mask value and also defer further
2881 		 * occurrences of this signal.
2882 		 *
2883 		 * Special case: user has done a sigsuspend.  Here the
2884 		 * current mask is not of interest, but rather the
2885 		 * mask from before the sigsuspend is what we want
2886 		 * restored after the signal processing is completed.
2887 		 */
2888 		if (td->td_pflags & TDP_OLDMASK) {
2889 			returnmask = td->td_oldsigmask;
2890 			td->td_pflags &= ~TDP_OLDMASK;
2891 		} else
2892 			returnmask = td->td_sigmask;
2893 
2894 		if (p->p_sig == sig) {
2895 			p->p_code = 0;
2896 			p->p_sig = 0;
2897 		}
2898 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
2899 		postsig_done(sig, td, ps);
2900 	}
2901 	return (1);
2902 }
2903 
2904 /*
2905  * Kill the current process for stated reason.
2906  */
2907 void
2908 killproc(p, why)
2909 	struct proc *p;
2910 	char *why;
2911 {
2912 
2913 	PROC_LOCK_ASSERT(p, MA_OWNED);
2914 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
2915 	    p->p_comm);
2916 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n", p->p_pid,
2917 	    p->p_comm, p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2918 	p->p_flag |= P_WKILLED;
2919 	kern_psignal(p, SIGKILL);
2920 }
2921 
2922 /*
2923  * Force the current process to exit with the specified signal, dumping core
2924  * if appropriate.  We bypass the normal tests for masked and caught signals,
2925  * allowing unrecoverable failures to terminate the process without changing
2926  * signal state.  Mark the accounting record with the signal termination.
2927  * If dumping core, save the signal number for the debugger.  Calls exit and
2928  * does not return.
2929  */
2930 void
2931 sigexit(td, sig)
2932 	struct thread *td;
2933 	int sig;
2934 {
2935 	struct proc *p = td->td_proc;
2936 
2937 	PROC_LOCK_ASSERT(p, MA_OWNED);
2938 	p->p_acflag |= AXSIG;
2939 	/*
2940 	 * We must be single-threading to generate a core dump.  This
2941 	 * ensures that the registers in the core file are up-to-date.
2942 	 * Also, the ELF dump handler assumes that the thread list doesn't
2943 	 * change out from under it.
2944 	 *
2945 	 * XXX If another thread attempts to single-thread before us
2946 	 *     (e.g. via fork()), we won't get a dump at all.
2947 	 */
2948 	if ((sigprop(sig) & SA_CORE) && thread_single(p, SINGLE_NO_EXIT) == 0) {
2949 		p->p_sig = sig;
2950 		/*
2951 		 * Log signals which would cause core dumps
2952 		 * (Log as LOG_INFO to appease those who don't want
2953 		 * these messages.)
2954 		 * XXX : Todo, as well as euid, write out ruid too
2955 		 * Note that coredump() drops proc lock.
2956 		 */
2957 		if (coredump(td) == 0)
2958 			sig |= WCOREFLAG;
2959 		if (kern_logsigexit)
2960 			log(LOG_INFO,
2961 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2962 			    p->p_pid, p->p_comm,
2963 			    td->td_ucred ? td->td_ucred->cr_uid : -1,
2964 			    sig &~ WCOREFLAG,
2965 			    sig & WCOREFLAG ? " (core dumped)" : "");
2966 	} else
2967 		PROC_UNLOCK(p);
2968 	exit1(td, 0, sig);
2969 	/* NOTREACHED */
2970 }
2971 
2972 /*
2973  * Send queued SIGCHLD to parent when child process's state
2974  * is changed.
2975  */
2976 static void
2977 sigparent(struct proc *p, int reason, int status)
2978 {
2979 	PROC_LOCK_ASSERT(p, MA_OWNED);
2980 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
2981 
2982 	if (p->p_ksi != NULL) {
2983 		p->p_ksi->ksi_signo  = SIGCHLD;
2984 		p->p_ksi->ksi_code   = reason;
2985 		p->p_ksi->ksi_status = status;
2986 		p->p_ksi->ksi_pid    = p->p_pid;
2987 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
2988 		if (KSI_ONQ(p->p_ksi))
2989 			return;
2990 	}
2991 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
2992 }
2993 
2994 static void
2995 childproc_jobstate(struct proc *p, int reason, int sig)
2996 {
2997 	struct sigacts *ps;
2998 
2999 	PROC_LOCK_ASSERT(p, MA_OWNED);
3000 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3001 
3002 	/*
3003 	 * Wake up parent sleeping in kern_wait(), also send
3004 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3005 	 * that parent will awake, because parent may masked
3006 	 * the signal.
3007 	 */
3008 	p->p_pptr->p_flag |= P_STATCHILD;
3009 	wakeup(p->p_pptr);
3010 
3011 	ps = p->p_pptr->p_sigacts;
3012 	mtx_lock(&ps->ps_mtx);
3013 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3014 		mtx_unlock(&ps->ps_mtx);
3015 		sigparent(p, reason, sig);
3016 	} else
3017 		mtx_unlock(&ps->ps_mtx);
3018 }
3019 
3020 void
3021 childproc_stopped(struct proc *p, int reason)
3022 {
3023 
3024 	childproc_jobstate(p, reason, p->p_xsig);
3025 }
3026 
3027 void
3028 childproc_continued(struct proc *p)
3029 {
3030 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3031 }
3032 
3033 void
3034 childproc_exited(struct proc *p)
3035 {
3036 	int reason, status;
3037 
3038 	if (WCOREDUMP(p->p_xsig)) {
3039 		reason = CLD_DUMPED;
3040 		status = WTERMSIG(p->p_xsig);
3041 	} else if (WIFSIGNALED(p->p_xsig)) {
3042 		reason = CLD_KILLED;
3043 		status = WTERMSIG(p->p_xsig);
3044 	} else {
3045 		reason = CLD_EXITED;
3046 		status = p->p_xexit;
3047 	}
3048 	/*
3049 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3050 	 * done in exit1().
3051 	 */
3052 	sigparent(p, reason, status);
3053 }
3054 
3055 /*
3056  * We only have 1 character for the core count in the format
3057  * string, so the range will be 0-9
3058  */
3059 #define MAX_NUM_CORES 10
3060 static int num_cores = 5;
3061 
3062 static int
3063 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3064 {
3065 	int error;
3066 	int new_val;
3067 
3068 	new_val = num_cores;
3069 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3070 	if (error != 0 || req->newptr == NULL)
3071 		return (error);
3072 	if (new_val > MAX_NUM_CORES)
3073 		new_val = MAX_NUM_CORES;
3074 	if (new_val < 0)
3075 		new_val = 0;
3076 	num_cores = new_val;
3077 	return (0);
3078 }
3079 SYSCTL_PROC(_debug, OID_AUTO, ncores, CTLTYPE_INT|CTLFLAG_RW,
3080 	    0, sizeof(int), sysctl_debug_num_cores_check, "I", "");
3081 
3082 #define	GZ_SUFFIX	".gz"
3083 
3084 #ifdef GZIO
3085 static int compress_user_cores = 1;
3086 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores, CTLFLAG_RWTUN,
3087     &compress_user_cores, 0, "Compression of user corefiles");
3088 
3089 int compress_user_cores_gzlevel = 6;
3090 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_gzlevel, CTLFLAG_RWTUN,
3091     &compress_user_cores_gzlevel, 0, "Corefile gzip compression level");
3092 #else
3093 static int compress_user_cores = 0;
3094 #endif
3095 
3096 /*
3097  * Protect the access to corefilename[] by allproc_lock.
3098  */
3099 #define	corefilename_lock	allproc_lock
3100 
3101 static char corefilename[MAXPATHLEN] = {"%N.core"};
3102 
3103 static int
3104 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3105 {
3106 	int error;
3107 
3108 	sx_xlock(&corefilename_lock);
3109 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3110 	    req);
3111 	sx_xunlock(&corefilename_lock);
3112 
3113 	return (error);
3114 }
3115 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RWTUN |
3116     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3117     "Process corefile name format string");
3118 
3119 /*
3120  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3121  * Expand the name described in corefilename, using name, uid, and pid
3122  * and open/create core file.
3123  * corefilename is a printf-like string, with three format specifiers:
3124  *	%N	name of process ("name")
3125  *	%P	process id (pid)
3126  *	%U	user id (uid)
3127  * For example, "%N.core" is the default; they can be disabled completely
3128  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3129  * This is controlled by the sysctl variable kern.corefile (see above).
3130  */
3131 static int
3132 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3133     int compress, struct vnode **vpp, char **namep)
3134 {
3135 	struct nameidata nd;
3136 	struct sbuf sb;
3137 	const char *format;
3138 	char *hostname, *name;
3139 	int indexpos, i, error, cmode, flags, oflags;
3140 
3141 	hostname = NULL;
3142 	format = corefilename;
3143 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3144 	indexpos = -1;
3145 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3146 	sx_slock(&corefilename_lock);
3147 	for (i = 0; format[i] != '\0'; i++) {
3148 		switch (format[i]) {
3149 		case '%':	/* Format character */
3150 			i++;
3151 			switch (format[i]) {
3152 			case '%':
3153 				sbuf_putc(&sb, '%');
3154 				break;
3155 			case 'H':	/* hostname */
3156 				if (hostname == NULL) {
3157 					hostname = malloc(MAXHOSTNAMELEN,
3158 					    M_TEMP, M_WAITOK);
3159 				}
3160 				getcredhostname(td->td_ucred, hostname,
3161 				    MAXHOSTNAMELEN);
3162 				sbuf_printf(&sb, "%s", hostname);
3163 				break;
3164 			case 'I':	/* autoincrementing index */
3165 				sbuf_printf(&sb, "0");
3166 				indexpos = sbuf_len(&sb) - 1;
3167 				break;
3168 			case 'N':	/* process name */
3169 				sbuf_printf(&sb, "%s", comm);
3170 				break;
3171 			case 'P':	/* process id */
3172 				sbuf_printf(&sb, "%u", pid);
3173 				break;
3174 			case 'U':	/* user id */
3175 				sbuf_printf(&sb, "%u", uid);
3176 				break;
3177 			default:
3178 				log(LOG_ERR,
3179 				    "Unknown format character %c in "
3180 				    "corename `%s'\n", format[i], format);
3181 				break;
3182 			}
3183 			break;
3184 		default:
3185 			sbuf_putc(&sb, format[i]);
3186 			break;
3187 		}
3188 	}
3189 	sx_sunlock(&corefilename_lock);
3190 	free(hostname, M_TEMP);
3191 	if (compress)
3192 		sbuf_printf(&sb, GZ_SUFFIX);
3193 	if (sbuf_error(&sb) != 0) {
3194 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3195 		    "long\n", (long)pid, comm, (u_long)uid);
3196 		sbuf_delete(&sb);
3197 		free(name, M_TEMP);
3198 		return (ENOMEM);
3199 	}
3200 	sbuf_finish(&sb);
3201 	sbuf_delete(&sb);
3202 
3203 	cmode = S_IRUSR | S_IWUSR;
3204 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3205 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3206 
3207 	/*
3208 	 * If the core format has a %I in it, then we need to check
3209 	 * for existing corefiles before returning a name.
3210 	 * To do this we iterate over 0..num_cores to find a
3211 	 * non-existing core file name to use.
3212 	 */
3213 	if (indexpos != -1) {
3214 		for (i = 0; i < num_cores; i++) {
3215 			flags = O_CREAT | O_EXCL | FWRITE | O_NOFOLLOW;
3216 			name[indexpos] = '0' + i;
3217 			NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3218 			error = vn_open_cred(&nd, &flags, cmode, oflags,
3219 			    td->td_ucred, NULL);
3220 			if (error) {
3221 				if (error == EEXIST)
3222 					continue;
3223 				log(LOG_ERR,
3224 				    "pid %d (%s), uid (%u):  Path `%s' failed "
3225 				    "on initial open test, error = %d\n",
3226 				    pid, comm, uid, name, error);
3227 			}
3228 			goto out;
3229 		}
3230 	}
3231 
3232 	flags = O_CREAT | FWRITE | O_NOFOLLOW;
3233 	NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3234 	error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred, NULL);
3235 out:
3236 	if (error) {
3237 #ifdef AUDIT
3238 		audit_proc_coredump(td, name, error);
3239 #endif
3240 		free(name, M_TEMP);
3241 		return (error);
3242 	}
3243 	NDFREE(&nd, NDF_ONLY_PNBUF);
3244 	*vpp = nd.ni_vp;
3245 	*namep = name;
3246 	return (0);
3247 }
3248 
3249 static int
3250 coredump_sanitise_path(const char *path)
3251 {
3252 	size_t i;
3253 
3254 	/*
3255 	 * Only send a subset of ASCII to devd(8) because it
3256 	 * might pass these strings to sh -c.
3257 	 */
3258 	for (i = 0; path[i]; i++)
3259 		if (!(isalpha(path[i]) || isdigit(path[i])) &&
3260 		    path[i] != '/' && path[i] != '.' &&
3261 		    path[i] != '-')
3262 			return (0);
3263 
3264 	return (1);
3265 }
3266 
3267 /*
3268  * Dump a process' core.  The main routine does some
3269  * policy checking, and creates the name of the coredump;
3270  * then it passes on a vnode and a size limit to the process-specific
3271  * coredump routine if there is one; if there _is not_ one, it returns
3272  * ENOSYS; otherwise it returns the error from the process-specific routine.
3273  */
3274 
3275 static int
3276 coredump(struct thread *td)
3277 {
3278 	struct proc *p = td->td_proc;
3279 	struct ucred *cred = td->td_ucred;
3280 	struct vnode *vp;
3281 	struct flock lf;
3282 	struct vattr vattr;
3283 	int error, error1, locked;
3284 	char *name;			/* name of corefile */
3285 	void *rl_cookie;
3286 	off_t limit;
3287 	char *data = NULL;
3288 	char *fullpath, *freepath = NULL;
3289 	size_t len;
3290 	static const char comm_name[] = "comm=";
3291 	static const char core_name[] = "core=";
3292 
3293 	PROC_LOCK_ASSERT(p, MA_OWNED);
3294 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3295 	_STOPEVENT(p, S_CORE, 0);
3296 
3297 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
3298 	    (p->p_flag2 & P2_NOTRACE) != 0) {
3299 		PROC_UNLOCK(p);
3300 		return (EFAULT);
3301 	}
3302 
3303 	/*
3304 	 * Note that the bulk of limit checking is done after
3305 	 * the corefile is created.  The exception is if the limit
3306 	 * for corefiles is 0, in which case we don't bother
3307 	 * creating the corefile at all.  This layout means that
3308 	 * a corefile is truncated instead of not being created,
3309 	 * if it is larger than the limit.
3310 	 */
3311 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
3312 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
3313 		PROC_UNLOCK(p);
3314 		return (EFBIG);
3315 	}
3316 	PROC_UNLOCK(p);
3317 
3318 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
3319 	    compress_user_cores, &vp, &name);
3320 	if (error != 0)
3321 		return (error);
3322 
3323 	/*
3324 	 * Don't dump to non-regular files or files with links.
3325 	 * Do not dump into system files.
3326 	 */
3327 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
3328 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0) {
3329 		VOP_UNLOCK(vp, 0);
3330 		error = EFAULT;
3331 		goto out;
3332 	}
3333 
3334 	VOP_UNLOCK(vp, 0);
3335 
3336 	/* Postpone other writers, including core dumps of other processes. */
3337 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
3338 
3339 	lf.l_whence = SEEK_SET;
3340 	lf.l_start = 0;
3341 	lf.l_len = 0;
3342 	lf.l_type = F_WRLCK;
3343 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
3344 
3345 	VATTR_NULL(&vattr);
3346 	vattr.va_size = 0;
3347 	if (set_core_nodump_flag)
3348 		vattr.va_flags = UF_NODUMP;
3349 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3350 	VOP_SETATTR(vp, &vattr, cred);
3351 	VOP_UNLOCK(vp, 0);
3352 	PROC_LOCK(p);
3353 	p->p_acflag |= ACORE;
3354 	PROC_UNLOCK(p);
3355 
3356 	if (p->p_sysent->sv_coredump != NULL) {
3357 		error = p->p_sysent->sv_coredump(td, vp, limit,
3358 		    compress_user_cores ? IMGACT_CORE_COMPRESS : 0);
3359 	} else {
3360 		error = ENOSYS;
3361 	}
3362 
3363 	if (locked) {
3364 		lf.l_type = F_UNLCK;
3365 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3366 	}
3367 	vn_rangelock_unlock(vp, rl_cookie);
3368 
3369 	/*
3370 	 * Notify the userland helper that a process triggered a core dump.
3371 	 * This allows the helper to run an automated debugging session.
3372 	 */
3373 	if (error != 0 || coredump_devctl == 0)
3374 		goto out;
3375 	len = MAXPATHLEN * 2 + sizeof(comm_name) - 1 +
3376 	    sizeof(' ') + sizeof(core_name) - 1;
3377 	data = malloc(len, M_TEMP, M_WAITOK);
3378 	if (vn_fullpath_global(td, p->p_textvp, &fullpath, &freepath) != 0)
3379 		goto out;
3380 	if (!coredump_sanitise_path(fullpath))
3381 		goto out;
3382 	snprintf(data, len, "%s%s ", comm_name, fullpath);
3383 	free(freepath, M_TEMP);
3384 	freepath = NULL;
3385 	if (vn_fullpath_global(td, vp, &fullpath, &freepath) != 0)
3386 		goto out;
3387 	if (!coredump_sanitise_path(fullpath))
3388 		goto out;
3389 	strlcat(data, core_name, len);
3390 	strlcat(data, fullpath, len);
3391 	devctl_notify("kernel", "signal", "coredump", data);
3392 out:
3393 	error1 = vn_close(vp, FWRITE, cred, td);
3394 	if (error == 0)
3395 		error = error1;
3396 #ifdef AUDIT
3397 	audit_proc_coredump(td, name, error);
3398 #endif
3399 	free(freepath, M_TEMP);
3400 	free(data, M_TEMP);
3401 	free(name, M_TEMP);
3402 	return (error);
3403 }
3404 
3405 /*
3406  * Nonexistent system call-- signal process (may want to handle it).  Flag
3407  * error in case process won't see signal immediately (blocked or ignored).
3408  */
3409 #ifndef _SYS_SYSPROTO_H_
3410 struct nosys_args {
3411 	int	dummy;
3412 };
3413 #endif
3414 /* ARGSUSED */
3415 int
3416 nosys(td, args)
3417 	struct thread *td;
3418 	struct nosys_args *args;
3419 {
3420 	struct proc *p = td->td_proc;
3421 
3422 	PROC_LOCK(p);
3423 	tdsignal(td, SIGSYS);
3424 	PROC_UNLOCK(p);
3425 	return (ENOSYS);
3426 }
3427 
3428 /*
3429  * Send a SIGIO or SIGURG signal to a process or process group using stored
3430  * credentials rather than those of the current process.
3431  */
3432 void
3433 pgsigio(sigiop, sig, checkctty)
3434 	struct sigio **sigiop;
3435 	int sig, checkctty;
3436 {
3437 	ksiginfo_t ksi;
3438 	struct sigio *sigio;
3439 
3440 	ksiginfo_init(&ksi);
3441 	ksi.ksi_signo = sig;
3442 	ksi.ksi_code = SI_KERNEL;
3443 
3444 	SIGIO_LOCK();
3445 	sigio = *sigiop;
3446 	if (sigio == NULL) {
3447 		SIGIO_UNLOCK();
3448 		return;
3449 	}
3450 	if (sigio->sio_pgid > 0) {
3451 		PROC_LOCK(sigio->sio_proc);
3452 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
3453 			kern_psignal(sigio->sio_proc, sig);
3454 		PROC_UNLOCK(sigio->sio_proc);
3455 	} else if (sigio->sio_pgid < 0) {
3456 		struct proc *p;
3457 
3458 		PGRP_LOCK(sigio->sio_pgrp);
3459 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
3460 			PROC_LOCK(p);
3461 			if (p->p_state == PRS_NORMAL &&
3462 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
3463 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
3464 				kern_psignal(p, sig);
3465 			PROC_UNLOCK(p);
3466 		}
3467 		PGRP_UNLOCK(sigio->sio_pgrp);
3468 	}
3469 	SIGIO_UNLOCK();
3470 }
3471 
3472 static int
3473 filt_sigattach(struct knote *kn)
3474 {
3475 	struct proc *p = curproc;
3476 
3477 	kn->kn_ptr.p_proc = p;
3478 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
3479 
3480 	knlist_add(&p->p_klist, kn, 0);
3481 
3482 	return (0);
3483 }
3484 
3485 static void
3486 filt_sigdetach(struct knote *kn)
3487 {
3488 	struct proc *p = kn->kn_ptr.p_proc;
3489 
3490 	knlist_remove(&p->p_klist, kn, 0);
3491 }
3492 
3493 /*
3494  * signal knotes are shared with proc knotes, so we apply a mask to
3495  * the hint in order to differentiate them from process hints.  This
3496  * could be avoided by using a signal-specific knote list, but probably
3497  * isn't worth the trouble.
3498  */
3499 static int
3500 filt_signal(struct knote *kn, long hint)
3501 {
3502 
3503 	if (hint & NOTE_SIGNAL) {
3504 		hint &= ~NOTE_SIGNAL;
3505 
3506 		if (kn->kn_id == hint)
3507 			kn->kn_data++;
3508 	}
3509 	return (kn->kn_data != 0);
3510 }
3511 
3512 struct sigacts *
3513 sigacts_alloc(void)
3514 {
3515 	struct sigacts *ps;
3516 
3517 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
3518 	refcount_init(&ps->ps_refcnt, 1);
3519 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
3520 	return (ps);
3521 }
3522 
3523 void
3524 sigacts_free(struct sigacts *ps)
3525 {
3526 
3527 	if (refcount_release(&ps->ps_refcnt) == 0)
3528 		return;
3529 	mtx_destroy(&ps->ps_mtx);
3530 	free(ps, M_SUBPROC);
3531 }
3532 
3533 struct sigacts *
3534 sigacts_hold(struct sigacts *ps)
3535 {
3536 
3537 	refcount_acquire(&ps->ps_refcnt);
3538 	return (ps);
3539 }
3540 
3541 void
3542 sigacts_copy(struct sigacts *dest, struct sigacts *src)
3543 {
3544 
3545 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
3546 	mtx_lock(&src->ps_mtx);
3547 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
3548 	mtx_unlock(&src->ps_mtx);
3549 }
3550 
3551 int
3552 sigacts_shared(struct sigacts *ps)
3553 {
3554 
3555 	return (ps->ps_refcnt > 1);
3556 }
3557